diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-1632/chat_template.jinja b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-1632/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..0ef09f214eaa6d9bca297988afc1454b5827b2c7
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-1632/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is true %}
+        {{- '<think>\n' }}
+    {%- else %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-1632/tokenizer_config.json b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-1632/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-1632/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-1632/trainer_state.json b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-1632/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..23f8e12159989457331b8fbf9b645e798dd22374
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-1632/trainer_state.json
@@ -0,0 +1,442 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 4.0,
+  "eval_steps": 500,
+  "global_step": 1632,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.2408065378665922,
+      "epoch": 0.12269938650306748,
+      "grad_norm": 1.6718658208847046,
+      "learning_rate": 4.487804970912036e-05,
+      "loss": 2.129624786376953,
+      "mean_token_accuracy": 0.5743644836544991,
+      "num_tokens": 134329.0,
+      "step": 50
+    },
+    {
+      "entropy": 1.1078504550457,
+      "epoch": 0.24539877300613497,
+      "grad_norm": 2.4091813564300537,
+      "learning_rate": 9.067197798373296e-05,
+      "loss": 1.0346390533447265,
+      "mean_token_accuracy": 0.7085381114482879,
+      "num_tokens": 262407.0,
+      "step": 100
+    },
+    {
+      "entropy": 0.9421598988771439,
+      "epoch": 0.36809815950920244,
+      "grad_norm": 1.3913646936416626,
+      "learning_rate": 0.00013646590625834558,
+      "loss": 0.8716551208496094,
+      "mean_token_accuracy": 0.7449040985107422,
+      "num_tokens": 401136.0,
+      "step": 150
+    },
+    {
+      "entropy": 0.8723550146818161,
+      "epoch": 0.49079754601226994,
+      "grad_norm": 1.1664468050003052,
+      "learning_rate": 0.0001822598345329582,
+      "loss": 0.8095977783203125,
+      "mean_token_accuracy": 0.7585240352153778,
+      "num_tokens": 531647.0,
+      "step": 200
+    },
+    {
+      "entropy": 0.8466519421339035,
+      "epoch": 0.6134969325153374,
+      "grad_norm": 1.0145421028137207,
+      "learning_rate": 0.00022805376280757083,
+      "loss": 0.7830857849121093,
+      "mean_token_accuracy": 0.7661705195903779,
+      "num_tokens": 666127.0,
+      "step": 250
+    },
+    {
+      "entropy": 0.8180295366048813,
+      "epoch": 0.7361963190184049,
+      "grad_norm": 0.9731705784797668,
+      "learning_rate": 0.00027384769108218336,
+      "loss": 0.7573603820800782,
+      "mean_token_accuracy": 0.7704453605413437,
+      "num_tokens": 799292.0,
+      "step": 300
+    },
+    {
+      "entropy": 0.7990227991342544,
+      "epoch": 0.8588957055214724,
+      "grad_norm": 1.0642374753952026,
+      "learning_rate": 0.000319641619356796,
+      "loss": 0.73709228515625,
+      "mean_token_accuracy": 0.7752733880281448,
+      "num_tokens": 930614.0,
+      "step": 350
+    },
+    {
+      "entropy": 0.7899087685346603,
+      "epoch": 0.9815950920245399,
+      "grad_norm": 0.880613386631012,
+      "learning_rate": 0.0003654355476314086,
+      "loss": 0.7347319793701171,
+      "mean_token_accuracy": 0.7766731631755829,
+      "num_tokens": 1061654.0,
+      "step": 400
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.83522472347532,
+      "eval_mean_token_accuracy": 0.7484098076820374,
+      "eval_not_syn_loss": 0.7872886061668396,
+      "eval_not_syn_runtime": 54.8697,
+      "eval_not_syn_samples_per_second": 25.424,
+      "eval_not_syn_steps_per_second": 3.189,
+      "eval_num_tokens": 1080196.0,
+      "step": 408
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.7784442271505083,
+      "eval_mean_token_accuracy": 0.8021791185651507,
+      "eval_num_tokens": 1080196.0,
+      "eval_syn_loss": 0.6980098485946655,
+      "eval_syn_runtime": 56.6342,
+      "eval_syn_samples_per_second": 24.632,
+      "eval_syn_steps_per_second": 3.09,
+      "step": 408
+    },
+    {
+      "entropy": 0.75672573694075,
+      "epoch": 1.1030674846625768,
+      "grad_norm": 0.9244198203086853,
+      "learning_rate": 0.00037356351883456205,
+      "loss": 0.7048745727539063,
+      "mean_token_accuracy": 0.783016896609104,
+      "num_tokens": 1199170.0,
+      "step": 450
+    },
+    {
+      "entropy": 0.7406797379255294,
+      "epoch": 1.2257668711656442,
+      "grad_norm": 0.6744896769523621,
+      "learning_rate": 0.00037311248151765587,
+      "loss": 0.6860730743408203,
+      "mean_token_accuracy": 0.787369327545166,
+      "num_tokens": 1334298.0,
+      "step": 500
+    },
+    {
+      "entropy": 0.7424973237514496,
+      "epoch": 1.3484662576687116,
+      "grad_norm": 0.92305588722229,
+      "learning_rate": 0.000372320629215228,
+      "loss": 0.6855724334716797,
+      "mean_token_accuracy": 0.7871134179830551,
+      "num_tokens": 1463189.0,
+      "step": 550
+    },
+    {
+      "entropy": 0.7502243828773498,
+      "epoch": 1.471165644171779,
+      "grad_norm": 0.8259047269821167,
+      "learning_rate": 0.00037118941074037944,
+      "loss": 0.6949834442138672,
+      "mean_token_accuracy": 0.7851914083957672,
+      "num_tokens": 1590310.0,
+      "step": 600
+    },
+    {
+      "entropy": 0.7233260518312454,
+      "epoch": 1.5938650306748468,
+      "grad_norm": 0.882722795009613,
+      "learning_rate": 0.00036972089582775814,
+      "loss": 0.6649005889892579,
+      "mean_token_accuracy": 0.7898879665136337,
+      "num_tokens": 1721836.0,
+      "step": 650
+    },
+    {
+      "entropy": 0.7162708967924118,
+      "epoch": 1.716564417177914,
+      "grad_norm": 0.6593027114868164,
+      "learning_rate": 0.0003679177713466678,
+      "loss": 0.6653710174560546,
+      "mean_token_accuracy": 0.7905678844451904,
+      "num_tokens": 1853507.0,
+      "step": 700
+    },
+    {
+      "entropy": 0.7150520020723343,
+      "epoch": 1.8392638036809816,
+      "grad_norm": 0.665340781211853,
+      "learning_rate": 0.0003657833363850354,
+      "loss": 0.6600938415527344,
+      "mean_token_accuracy": 0.7918649202585221,
+      "num_tokens": 1988113.0,
+      "step": 750
+    },
+    {
+      "entropy": 0.6993386310338974,
+      "epoch": 1.961963190184049,
+      "grad_norm": 0.6502187252044678,
+      "learning_rate": 0.00036332149621323294,
+      "loss": 0.6469607543945313,
+      "mean_token_accuracy": 0.7950778317451477,
+      "num_tokens": 2121314.0,
+      "step": 800
+    },
+    {
+      "epoch": 2.0,
+      "eval_entropy": 0.7315139893123082,
+      "eval_mean_token_accuracy": 0.7713504004478454,
+      "eval_not_syn_loss": 0.699893593788147,
+      "eval_not_syn_runtime": 55.4135,
+      "eval_not_syn_samples_per_second": 25.174,
+      "eval_not_syn_steps_per_second": 3.158,
+      "eval_num_tokens": 2160392.0,
+      "step": 816
+    },
+    {
+      "epoch": 2.0,
+      "eval_entropy": 0.6848451798302787,
+      "eval_mean_token_accuracy": 0.814895657471248,
+      "eval_num_tokens": 2160392.0,
+      "eval_syn_loss": 0.6415576934814453,
+      "eval_syn_runtime": 56.4735,
+      "eval_syn_samples_per_second": 24.702,
+      "eval_syn_steps_per_second": 3.099,
+      "step": 816
+    },
+    {
+      "entropy": 0.6596363644407253,
+      "epoch": 2.083435582822086,
+      "grad_norm": 0.6647700071334839,
+      "learning_rate": 0.00036053675513879666,
+      "loss": 0.599151611328125,
+      "mean_token_accuracy": 0.8071986864311527,
+      "num_tokens": 2253179.0,
+      "step": 850
+    },
+    {
+      "entropy": 0.6626429110765457,
+      "epoch": 2.2061349693251535,
+      "grad_norm": 0.7356868386268616,
+      "learning_rate": 0.00035743420826511783,
+      "loss": 0.6004195404052735,
+      "mean_token_accuracy": 0.805362731218338,
+      "num_tokens": 2379965.0,
+      "step": 900
+    },
+    {
+      "entropy": 0.6503470712900161,
+      "epoch": 2.3288343558282207,
+      "grad_norm": 1.095608115196228,
+      "learning_rate": 0.0003540195321691835,
+      "loss": 0.5951333999633789,
+      "mean_token_accuracy": 0.807324025630951,
+      "num_tokens": 2515593.0,
+      "step": 950
+    },
+    {
+      "entropy": 0.660577870607376,
+      "epoch": 2.4515337423312884,
+      "grad_norm": 0.6602179408073425,
+      "learning_rate": 0.00035029897451542325,
+      "loss": 0.6056365585327148,
+      "mean_token_accuracy": 0.8057438349723816,
+      "num_tokens": 2642790.0,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6555853113532066,
+      "epoch": 2.574233128834356,
+      "grad_norm": 0.7459151148796082,
+      "learning_rate": 0.00034627934262466606,
+      "loss": 0.5963204193115235,
+      "mean_token_accuracy": 0.8060598099231719,
+      "num_tokens": 2774396.0,
+      "step": 1050
+    },
+    {
+      "entropy": 0.6505853188037872,
+      "epoch": 2.6969325153374233,
+      "grad_norm": 0.6359792947769165,
+      "learning_rate": 0.00034196799101912103,
+      "loss": 0.5993065643310547,
+      "mean_token_accuracy": 0.806280387043953,
+      "num_tokens": 2912102.0,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6509970253705979,
+      "epoch": 2.819631901840491,
+      "grad_norm": 0.8086174726486206,
+      "learning_rate": 0.0003373728079661707,
+      "loss": 0.6001268005371094,
+      "mean_token_accuracy": 0.806866540312767,
+      "num_tokens": 3042987.0,
+      "step": 1150
+    },
+    {
+      "entropy": 0.6404844307899475,
+      "epoch": 2.942331288343558,
+      "grad_norm": 0.7038407325744629,
+      "learning_rate": 0.0003325022010455975,
+      "loss": 0.5870526123046875,
+      "mean_token_accuracy": 0.8104586935043335,
+      "num_tokens": 3178044.0,
+      "step": 1200
+    },
+    {
+      "epoch": 3.0,
+      "eval_entropy": 0.6645996727262224,
+      "eval_mean_token_accuracy": 0.798654237134116,
+      "eval_not_syn_loss": 0.6586771011352539,
+      "eval_not_syn_runtime": 55.1088,
+      "eval_not_syn_samples_per_second": 25.314,
+      "eval_not_syn_steps_per_second": 3.176,
+      "eval_num_tokens": 3240588.0,
+      "step": 1224
+    },
+    {
+      "epoch": 3.0,
+      "eval_entropy": 0.6233607809884207,
+      "eval_mean_token_accuracy": 0.7977249867575509,
+      "eval_num_tokens": 3240588.0,
+      "eval_syn_loss": 0.6288298964500427,
+      "eval_syn_runtime": 56.6081,
+      "eval_syn_samples_per_second": 24.643,
+      "eval_syn_steps_per_second": 3.091,
+      "step": 1224
+    },
+    {
+      "entropy": 0.6081961104364106,
+      "epoch": 3.063803680981595,
+      "grad_norm": 0.6161203980445862,
+      "learning_rate": 0.0003273650817666497,
+      "loss": 0.5481716537475586,
+      "mean_token_accuracy": 0.8175542035488167,
+      "num_tokens": 3312797.0,
+      "step": 1250
+    },
+    {
+      "entropy": 0.5877064895629883,
+      "epoch": 3.1865030674846624,
+      "grad_norm": 0.6909636855125427,
+      "learning_rate": 0.00032197084926309234,
+      "loss": 0.5224150466918945,
+      "mean_token_accuracy": 0.8250984263420105,
+      "num_tokens": 3441629.0,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5952656552195549,
+      "epoch": 3.30920245398773,
+      "grad_norm": 0.8147026300430298,
+      "learning_rate": 0.0003163293730960758,
+      "loss": 0.5349474334716797,
+      "mean_token_accuracy": 0.8213064837455749,
+      "num_tokens": 3576093.0,
+      "step": 1350
+    },
+    {
+      "entropy": 0.5937123569846153,
+      "epoch": 3.4319018404907977,
+      "grad_norm": 0.8690970540046692,
+      "learning_rate": 0.0003104509751962869,
+      "loss": 0.5310481262207031,
+      "mean_token_accuracy": 0.8229135835170746,
+      "num_tokens": 3705912.0,
+      "step": 1400
+    },
+    {
+      "entropy": 0.6067730465531349,
+      "epoch": 3.554601226993865,
+      "grad_norm": 0.7433330416679382,
+      "learning_rate": 0.0003043464109784207,
+      "loss": 0.5450045394897461,
+      "mean_token_accuracy": 0.8205251175165177,
+      "num_tokens": 3832376.0,
+      "step": 1450
+    },
+    {
+      "entropy": 0.5926914718747139,
+      "epoch": 3.6773006134969326,
+      "grad_norm": 0.7265587449073792,
+      "learning_rate": 0.00029802684966252974,
+      "loss": 0.5330168914794922,
+      "mean_token_accuracy": 0.821294127702713,
+      "num_tokens": 3967273.0,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5881199997663498,
+      "epoch": 3.8,
+      "grad_norm": 0.7681817412376404,
+      "learning_rate": 0.00029150385383825187,
+      "loss": 0.5288929367065429,
+      "mean_token_accuracy": 0.8234161508083343,
+      "num_tokens": 4102434.0,
+      "step": 1550
+    },
+    {
+      "entropy": 0.6008686199784279,
+      "epoch": 3.9226993865030675,
+      "grad_norm": 0.9384056329727173,
+      "learning_rate": 0.0002847893583093113,
+      "loss": 0.5437511444091797,
+      "mean_token_accuracy": 0.8201366275548935,
+      "num_tokens": 4234037.0,
+      "step": 1600
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.6218148783275059,
+      "eval_mean_token_accuracy": 0.783796865599496,
+      "eval_not_syn_loss": 0.6558170318603516,
+      "eval_not_syn_runtime": 55.429,
+      "eval_not_syn_samples_per_second": 25.167,
+      "eval_not_syn_steps_per_second": 3.157,
+      "eval_num_tokens": 4320784.0,
+      "step": 1632
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5833754854542869,
+      "eval_mean_token_accuracy": 0.8278223422595433,
+      "eval_num_tokens": 4320784.0,
+      "eval_syn_loss": 0.604770302772522,
+      "eval_syn_runtime": 56.9593,
+      "eval_syn_samples_per_second": 24.491,
+      "eval_syn_steps_per_second": 3.072,
+      "step": 1632
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 4080,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 8.61560698569166e+16,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2040/README.md b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2040/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..7ac50e2706e5b4eae8f28e204601afb241c49e55
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2040/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-2B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-2B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2040/adapter_config.json b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2040/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b9596096c71017d49955136a4e23d9f97138d305
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2040/adapter_config.json
@@ -0,0 +1,48 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.08602048083239339,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "q_proj",
+    "k_proj",
+    "gate_proj",
+    "up_proj",
+    "o_proj",
+    "down_proj",
+    "v_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2040/chat_template.jinja b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2040/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..0ef09f214eaa6d9bca297988afc1454b5827b2c7
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2040/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is true %}
+        {{- '<think>\n' }}
+    {%- else %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2040/tokenizer_config.json b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2040/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2040/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2040/trainer_state.json b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2040/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..2fee862bcd9f592d79ab4664bfc26714e922470c
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2040/trainer_state.json
@@ -0,0 +1,544 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 5.0,
+  "eval_steps": 500,
+  "global_step": 2040,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.2408065378665922,
+      "epoch": 0.12269938650306748,
+      "grad_norm": 1.6718658208847046,
+      "learning_rate": 4.487804970912036e-05,
+      "loss": 2.129624786376953,
+      "mean_token_accuracy": 0.5743644836544991,
+      "num_tokens": 134329.0,
+      "step": 50
+    },
+    {
+      "entropy": 1.1078504550457,
+      "epoch": 0.24539877300613497,
+      "grad_norm": 2.4091813564300537,
+      "learning_rate": 9.067197798373296e-05,
+      "loss": 1.0346390533447265,
+      "mean_token_accuracy": 0.7085381114482879,
+      "num_tokens": 262407.0,
+      "step": 100
+    },
+    {
+      "entropy": 0.9421598988771439,
+      "epoch": 0.36809815950920244,
+      "grad_norm": 1.3913646936416626,
+      "learning_rate": 0.00013646590625834558,
+      "loss": 0.8716551208496094,
+      "mean_token_accuracy": 0.7449040985107422,
+      "num_tokens": 401136.0,
+      "step": 150
+    },
+    {
+      "entropy": 0.8723550146818161,
+      "epoch": 0.49079754601226994,
+      "grad_norm": 1.1664468050003052,
+      "learning_rate": 0.0001822598345329582,
+      "loss": 0.8095977783203125,
+      "mean_token_accuracy": 0.7585240352153778,
+      "num_tokens": 531647.0,
+      "step": 200
+    },
+    {
+      "entropy": 0.8466519421339035,
+      "epoch": 0.6134969325153374,
+      "grad_norm": 1.0145421028137207,
+      "learning_rate": 0.00022805376280757083,
+      "loss": 0.7830857849121093,
+      "mean_token_accuracy": 0.7661705195903779,
+      "num_tokens": 666127.0,
+      "step": 250
+    },
+    {
+      "entropy": 0.8180295366048813,
+      "epoch": 0.7361963190184049,
+      "grad_norm": 0.9731705784797668,
+      "learning_rate": 0.00027384769108218336,
+      "loss": 0.7573603820800782,
+      "mean_token_accuracy": 0.7704453605413437,
+      "num_tokens": 799292.0,
+      "step": 300
+    },
+    {
+      "entropy": 0.7990227991342544,
+      "epoch": 0.8588957055214724,
+      "grad_norm": 1.0642374753952026,
+      "learning_rate": 0.000319641619356796,
+      "loss": 0.73709228515625,
+      "mean_token_accuracy": 0.7752733880281448,
+      "num_tokens": 930614.0,
+      "step": 350
+    },
+    {
+      "entropy": 0.7899087685346603,
+      "epoch": 0.9815950920245399,
+      "grad_norm": 0.880613386631012,
+      "learning_rate": 0.0003654355476314086,
+      "loss": 0.7347319793701171,
+      "mean_token_accuracy": 0.7766731631755829,
+      "num_tokens": 1061654.0,
+      "step": 400
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.83522472347532,
+      "eval_mean_token_accuracy": 0.7484098076820374,
+      "eval_not_syn_loss": 0.7872886061668396,
+      "eval_not_syn_runtime": 54.8697,
+      "eval_not_syn_samples_per_second": 25.424,
+      "eval_not_syn_steps_per_second": 3.189,
+      "eval_num_tokens": 1080196.0,
+      "step": 408
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.7784442271505083,
+      "eval_mean_token_accuracy": 0.8021791185651507,
+      "eval_num_tokens": 1080196.0,
+      "eval_syn_loss": 0.6980098485946655,
+      "eval_syn_runtime": 56.6342,
+      "eval_syn_samples_per_second": 24.632,
+      "eval_syn_steps_per_second": 3.09,
+      "step": 408
+    },
+    {
+      "entropy": 0.75672573694075,
+      "epoch": 1.1030674846625768,
+      "grad_norm": 0.9244198203086853,
+      "learning_rate": 0.00037356351883456205,
+      "loss": 0.7048745727539063,
+      "mean_token_accuracy": 0.783016896609104,
+      "num_tokens": 1199170.0,
+      "step": 450
+    },
+    {
+      "entropy": 0.7406797379255294,
+      "epoch": 1.2257668711656442,
+      "grad_norm": 0.6744896769523621,
+      "learning_rate": 0.00037311248151765587,
+      "loss": 0.6860730743408203,
+      "mean_token_accuracy": 0.787369327545166,
+      "num_tokens": 1334298.0,
+      "step": 500
+    },
+    {
+      "entropy": 0.7424973237514496,
+      "epoch": 1.3484662576687116,
+      "grad_norm": 0.92305588722229,
+      "learning_rate": 0.000372320629215228,
+      "loss": 0.6855724334716797,
+      "mean_token_accuracy": 0.7871134179830551,
+      "num_tokens": 1463189.0,
+      "step": 550
+    },
+    {
+      "entropy": 0.7502243828773498,
+      "epoch": 1.471165644171779,
+      "grad_norm": 0.8259047269821167,
+      "learning_rate": 0.00037118941074037944,
+      "loss": 0.6949834442138672,
+      "mean_token_accuracy": 0.7851914083957672,
+      "num_tokens": 1590310.0,
+      "step": 600
+    },
+    {
+      "entropy": 0.7233260518312454,
+      "epoch": 1.5938650306748468,
+      "grad_norm": 0.882722795009613,
+      "learning_rate": 0.00036972089582775814,
+      "loss": 0.6649005889892579,
+      "mean_token_accuracy": 0.7898879665136337,
+      "num_tokens": 1721836.0,
+      "step": 650
+    },
+    {
+      "entropy": 0.7162708967924118,
+      "epoch": 1.716564417177914,
+      "grad_norm": 0.6593027114868164,
+      "learning_rate": 0.0003679177713466678,
+      "loss": 0.6653710174560546,
+      "mean_token_accuracy": 0.7905678844451904,
+      "num_tokens": 1853507.0,
+      "step": 700
+    },
+    {
+      "entropy": 0.7150520020723343,
+      "epoch": 1.8392638036809816,
+      "grad_norm": 0.665340781211853,
+      "learning_rate": 0.0003657833363850354,
+      "loss": 0.6600938415527344,
+      "mean_token_accuracy": 0.7918649202585221,
+      "num_tokens": 1988113.0,
+      "step": 750
+    },
+    {
+      "entropy": 0.6993386310338974,
+      "epoch": 1.961963190184049,
+      "grad_norm": 0.6502187252044678,
+      "learning_rate": 0.00036332149621323294,
+      "loss": 0.6469607543945313,
+      "mean_token_accuracy": 0.7950778317451477,
+      "num_tokens": 2121314.0,
+      "step": 800
+    },
+    {
+      "epoch": 2.0,
+      "eval_entropy": 0.7315139893123082,
+      "eval_mean_token_accuracy": 0.7713504004478454,
+      "eval_not_syn_loss": 0.699893593788147,
+      "eval_not_syn_runtime": 55.4135,
+      "eval_not_syn_samples_per_second": 25.174,
+      "eval_not_syn_steps_per_second": 3.158,
+      "eval_num_tokens": 2160392.0,
+      "step": 816
+    },
+    {
+      "epoch": 2.0,
+      "eval_entropy": 0.6848451798302787,
+      "eval_mean_token_accuracy": 0.814895657471248,
+      "eval_num_tokens": 2160392.0,
+      "eval_syn_loss": 0.6415576934814453,
+      "eval_syn_runtime": 56.4735,
+      "eval_syn_samples_per_second": 24.702,
+      "eval_syn_steps_per_second": 3.099,
+      "step": 816
+    },
+    {
+      "entropy": 0.6596363644407253,
+      "epoch": 2.083435582822086,
+      "grad_norm": 0.6647700071334839,
+      "learning_rate": 0.00036053675513879666,
+      "loss": 0.599151611328125,
+      "mean_token_accuracy": 0.8071986864311527,
+      "num_tokens": 2253179.0,
+      "step": 850
+    },
+    {
+      "entropy": 0.6626429110765457,
+      "epoch": 2.2061349693251535,
+      "grad_norm": 0.7356868386268616,
+      "learning_rate": 0.00035743420826511783,
+      "loss": 0.6004195404052735,
+      "mean_token_accuracy": 0.805362731218338,
+      "num_tokens": 2379965.0,
+      "step": 900
+    },
+    {
+      "entropy": 0.6503470712900161,
+      "epoch": 2.3288343558282207,
+      "grad_norm": 1.095608115196228,
+      "learning_rate": 0.0003540195321691835,
+      "loss": 0.5951333999633789,
+      "mean_token_accuracy": 0.807324025630951,
+      "num_tokens": 2515593.0,
+      "step": 950
+    },
+    {
+      "entropy": 0.660577870607376,
+      "epoch": 2.4515337423312884,
+      "grad_norm": 0.6602179408073425,
+      "learning_rate": 0.00035029897451542325,
+      "loss": 0.6056365585327148,
+      "mean_token_accuracy": 0.8057438349723816,
+      "num_tokens": 2642790.0,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6555853113532066,
+      "epoch": 2.574233128834356,
+      "grad_norm": 0.7459151148796082,
+      "learning_rate": 0.00034627934262466606,
+      "loss": 0.5963204193115235,
+      "mean_token_accuracy": 0.8060598099231719,
+      "num_tokens": 2774396.0,
+      "step": 1050
+    },
+    {
+      "entropy": 0.6505853188037872,
+      "epoch": 2.6969325153374233,
+      "grad_norm": 0.6359792947769165,
+      "learning_rate": 0.00034196799101912103,
+      "loss": 0.5993065643310547,
+      "mean_token_accuracy": 0.806280387043953,
+      "num_tokens": 2912102.0,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6509970253705979,
+      "epoch": 2.819631901840491,
+      "grad_norm": 0.8086174726486206,
+      "learning_rate": 0.0003373728079661707,
+      "loss": 0.6001268005371094,
+      "mean_token_accuracy": 0.806866540312767,
+      "num_tokens": 3042987.0,
+      "step": 1150
+    },
+    {
+      "entropy": 0.6404844307899475,
+      "epoch": 2.942331288343558,
+      "grad_norm": 0.7038407325744629,
+      "learning_rate": 0.0003325022010455975,
+      "loss": 0.5870526123046875,
+      "mean_token_accuracy": 0.8104586935043335,
+      "num_tokens": 3178044.0,
+      "step": 1200
+    },
+    {
+      "epoch": 3.0,
+      "eval_entropy": 0.6645996727262224,
+      "eval_mean_token_accuracy": 0.798654237134116,
+      "eval_not_syn_loss": 0.6586771011352539,
+      "eval_not_syn_runtime": 55.1088,
+      "eval_not_syn_samples_per_second": 25.314,
+      "eval_not_syn_steps_per_second": 3.176,
+      "eval_num_tokens": 3240588.0,
+      "step": 1224
+    },
+    {
+      "epoch": 3.0,
+      "eval_entropy": 0.6233607809884207,
+      "eval_mean_token_accuracy": 0.7977249867575509,
+      "eval_num_tokens": 3240588.0,
+      "eval_syn_loss": 0.6288298964500427,
+      "eval_syn_runtime": 56.6081,
+      "eval_syn_samples_per_second": 24.643,
+      "eval_syn_steps_per_second": 3.091,
+      "step": 1224
+    },
+    {
+      "entropy": 0.6081961104364106,
+      "epoch": 3.063803680981595,
+      "grad_norm": 0.6161203980445862,
+      "learning_rate": 0.0003273650817666497,
+      "loss": 0.5481716537475586,
+      "mean_token_accuracy": 0.8175542035488167,
+      "num_tokens": 3312797.0,
+      "step": 1250
+    },
+    {
+      "entropy": 0.5877064895629883,
+      "epoch": 3.1865030674846624,
+      "grad_norm": 0.6909636855125427,
+      "learning_rate": 0.00032197084926309234,
+      "loss": 0.5224150466918945,
+      "mean_token_accuracy": 0.8250984263420105,
+      "num_tokens": 3441629.0,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5952656552195549,
+      "epoch": 3.30920245398773,
+      "grad_norm": 0.8147026300430298,
+      "learning_rate": 0.0003163293730960758,
+      "loss": 0.5349474334716797,
+      "mean_token_accuracy": 0.8213064837455749,
+      "num_tokens": 3576093.0,
+      "step": 1350
+    },
+    {
+      "entropy": 0.5937123569846153,
+      "epoch": 3.4319018404907977,
+      "grad_norm": 0.8690970540046692,
+      "learning_rate": 0.0003104509751962869,
+      "loss": 0.5310481262207031,
+      "mean_token_accuracy": 0.8229135835170746,
+      "num_tokens": 3705912.0,
+      "step": 1400
+    },
+    {
+      "entropy": 0.6067730465531349,
+      "epoch": 3.554601226993865,
+      "grad_norm": 0.7433330416679382,
+      "learning_rate": 0.0003043464109784207,
+      "loss": 0.5450045394897461,
+      "mean_token_accuracy": 0.8205251175165177,
+      "num_tokens": 3832376.0,
+      "step": 1450
+    },
+    {
+      "entropy": 0.5926914718747139,
+      "epoch": 3.6773006134969326,
+      "grad_norm": 0.7265587449073792,
+      "learning_rate": 0.00029802684966252974,
+      "loss": 0.5330168914794922,
+      "mean_token_accuracy": 0.821294127702713,
+      "num_tokens": 3967273.0,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5881199997663498,
+      "epoch": 3.8,
+      "grad_norm": 0.7681817412376404,
+      "learning_rate": 0.00029150385383825187,
+      "loss": 0.5288929367065429,
+      "mean_token_accuracy": 0.8234161508083343,
+      "num_tokens": 4102434.0,
+      "step": 1550
+    },
+    {
+      "entropy": 0.6008686199784279,
+      "epoch": 3.9226993865030675,
+      "grad_norm": 0.9384056329727173,
+      "learning_rate": 0.0002847893583093113,
+      "loss": 0.5437511444091797,
+      "mean_token_accuracy": 0.8201366275548935,
+      "num_tokens": 4234037.0,
+      "step": 1600
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.6218148783275059,
+      "eval_mean_token_accuracy": 0.783796865599496,
+      "eval_not_syn_loss": 0.6558170318603516,
+      "eval_not_syn_runtime": 55.429,
+      "eval_not_syn_samples_per_second": 25.167,
+      "eval_not_syn_steps_per_second": 3.157,
+      "eval_num_tokens": 4320784.0,
+      "step": 1632
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5833754854542869,
+      "eval_mean_token_accuracy": 0.8278223422595433,
+      "eval_num_tokens": 4320784.0,
+      "eval_syn_loss": 0.604770302772522,
+      "eval_syn_runtime": 56.9593,
+      "eval_syn_samples_per_second": 24.491,
+      "eval_syn_steps_per_second": 3.072,
+      "step": 1632
+    },
+    {
+      "entropy": 0.559690699733869,
+      "epoch": 4.044171779141104,
+      "grad_norm": 0.65739905834198,
+      "learning_rate": 0.0002778956482569963,
+      "loss": 0.49702674865722657,
+      "mean_token_accuracy": 0.8317034166268628,
+      "num_tokens": 4368268.0,
+      "step": 1650
+    },
+    {
+      "entropy": 0.5401161375641823,
+      "epoch": 4.166871165644172,
+      "grad_norm": 1.5176353454589844,
+      "learning_rate": 0.0002708353367625691,
+      "loss": 0.4598746109008789,
+      "mean_token_accuracy": 0.8412033343315124,
+      "num_tokens": 4494525.0,
+      "step": 1700
+    },
+    {
+      "entropy": 0.5299916860461235,
+      "epoch": 4.289570552147239,
+      "grad_norm": 0.685965359210968,
+      "learning_rate": 0.0002636213417297333,
+      "loss": 0.45663665771484374,
+      "mean_token_accuracy": 0.8412413811683654,
+      "num_tokens": 4631985.0,
+      "step": 1750
+    },
+    {
+      "entropy": 0.5352088099718094,
+      "epoch": 4.412269938650307,
+      "grad_norm": 0.8471158742904663,
+      "learning_rate": 0.0002562668622493824,
+      "loss": 0.4625707626342773,
+      "mean_token_accuracy": 0.8400991547107697,
+      "num_tokens": 4766323.0,
+      "step": 1800
+    },
+    {
+      "entropy": 0.5425067806243896,
+      "epoch": 4.534969325153375,
+      "grad_norm": 0.8595437407493591,
+      "learning_rate": 0.00024878535444987474,
+      "loss": 0.47124954223632814,
+      "mean_token_accuracy": 0.8376108628511428,
+      "num_tokens": 4895606.0,
+      "step": 1850
+    },
+    {
+      "entropy": 0.5419312778115273,
+      "epoch": 4.6576687116564415,
+      "grad_norm": 0.8436959981918335,
+      "learning_rate": 0.00024119050687701818,
+      "loss": 0.4708108901977539,
+      "mean_token_accuracy": 0.8390797019004822,
+      "num_tokens": 5025912.0,
+      "step": 1900
+    },
+    {
+      "entropy": 0.5392271488904953,
+      "epoch": 4.780368098159509,
+      "grad_norm": 0.8691944479942322,
+      "learning_rate": 0.00023349621544881404,
+      "loss": 0.46682029724121094,
+      "mean_token_accuracy": 0.8385172665119172,
+      "num_tokens": 5162972.0,
+      "step": 1950
+    },
+    {
+      "entropy": 0.5332251599431038,
+      "epoch": 4.903067484662577,
+      "grad_norm": 0.6814598441123962,
+      "learning_rate": 0.00022571655803078078,
+      "loss": 0.4687493896484375,
+      "mean_token_accuracy": 0.8388861840963364,
+      "num_tokens": 5297935.0,
+      "step": 2000
+    },
+    {
+      "epoch": 5.0,
+      "eval_entropy": 0.5847777877535139,
+      "eval_mean_token_accuracy": 0.7782382583618164,
+      "eval_not_syn_loss": 0.6678160429000854,
+      "eval_not_syn_runtime": 55.1467,
+      "eval_not_syn_samples_per_second": 25.296,
+      "eval_not_syn_steps_per_second": 3.173,
+      "eval_num_tokens": 5400980.0,
+      "step": 2040
+    },
+    {
+      "epoch": 5.0,
+      "eval_entropy": 0.5505962218557086,
+      "eval_mean_token_accuracy": 0.8308270008223397,
+      "eval_num_tokens": 5400980.0,
+      "eval_syn_loss": 0.606867253780365,
+      "eval_syn_runtime": 56.9768,
+      "eval_syn_samples_per_second": 24.484,
+      "eval_syn_steps_per_second": 3.071,
+      "step": 2040
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 4080,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.0759747166537664e+17,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2448/README.md b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2448/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..7ac50e2706e5b4eae8f28e204601afb241c49e55
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2448/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-2B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-2B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2448/adapter_config.json b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2448/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b9596096c71017d49955136a4e23d9f97138d305
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2448/adapter_config.json
@@ -0,0 +1,48 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.08602048083239339,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "q_proj",
+    "k_proj",
+    "gate_proj",
+    "up_proj",
+    "o_proj",
+    "down_proj",
+    "v_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2448/chat_template.jinja b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2448/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..0ef09f214eaa6d9bca297988afc1454b5827b2c7
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2448/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is true %}
+        {{- '<think>\n' }}
+    {%- else %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2448/tokenizer_config.json b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2448/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2448/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2448/trainer_state.json b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2448/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..d6fa1e1c23b03a45c389f5c77af70f4697f74f13
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2448/trainer_state.json
@@ -0,0 +1,646 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 6.0,
+  "eval_steps": 500,
+  "global_step": 2448,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.2408065378665922,
+      "epoch": 0.12269938650306748,
+      "grad_norm": 1.6718658208847046,
+      "learning_rate": 4.487804970912036e-05,
+      "loss": 2.129624786376953,
+      "mean_token_accuracy": 0.5743644836544991,
+      "num_tokens": 134329.0,
+      "step": 50
+    },
+    {
+      "entropy": 1.1078504550457,
+      "epoch": 0.24539877300613497,
+      "grad_norm": 2.4091813564300537,
+      "learning_rate": 9.067197798373296e-05,
+      "loss": 1.0346390533447265,
+      "mean_token_accuracy": 0.7085381114482879,
+      "num_tokens": 262407.0,
+      "step": 100
+    },
+    {
+      "entropy": 0.9421598988771439,
+      "epoch": 0.36809815950920244,
+      "grad_norm": 1.3913646936416626,
+      "learning_rate": 0.00013646590625834558,
+      "loss": 0.8716551208496094,
+      "mean_token_accuracy": 0.7449040985107422,
+      "num_tokens": 401136.0,
+      "step": 150
+    },
+    {
+      "entropy": 0.8723550146818161,
+      "epoch": 0.49079754601226994,
+      "grad_norm": 1.1664468050003052,
+      "learning_rate": 0.0001822598345329582,
+      "loss": 0.8095977783203125,
+      "mean_token_accuracy": 0.7585240352153778,
+      "num_tokens": 531647.0,
+      "step": 200
+    },
+    {
+      "entropy": 0.8466519421339035,
+      "epoch": 0.6134969325153374,
+      "grad_norm": 1.0145421028137207,
+      "learning_rate": 0.00022805376280757083,
+      "loss": 0.7830857849121093,
+      "mean_token_accuracy": 0.7661705195903779,
+      "num_tokens": 666127.0,
+      "step": 250
+    },
+    {
+      "entropy": 0.8180295366048813,
+      "epoch": 0.7361963190184049,
+      "grad_norm": 0.9731705784797668,
+      "learning_rate": 0.00027384769108218336,
+      "loss": 0.7573603820800782,
+      "mean_token_accuracy": 0.7704453605413437,
+      "num_tokens": 799292.0,
+      "step": 300
+    },
+    {
+      "entropy": 0.7990227991342544,
+      "epoch": 0.8588957055214724,
+      "grad_norm": 1.0642374753952026,
+      "learning_rate": 0.000319641619356796,
+      "loss": 0.73709228515625,
+      "mean_token_accuracy": 0.7752733880281448,
+      "num_tokens": 930614.0,
+      "step": 350
+    },
+    {
+      "entropy": 0.7899087685346603,
+      "epoch": 0.9815950920245399,
+      "grad_norm": 0.880613386631012,
+      "learning_rate": 0.0003654355476314086,
+      "loss": 0.7347319793701171,
+      "mean_token_accuracy": 0.7766731631755829,
+      "num_tokens": 1061654.0,
+      "step": 400
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.83522472347532,
+      "eval_mean_token_accuracy": 0.7484098076820374,
+      "eval_not_syn_loss": 0.7872886061668396,
+      "eval_not_syn_runtime": 54.8697,
+      "eval_not_syn_samples_per_second": 25.424,
+      "eval_not_syn_steps_per_second": 3.189,
+      "eval_num_tokens": 1080196.0,
+      "step": 408
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.7784442271505083,
+      "eval_mean_token_accuracy": 0.8021791185651507,
+      "eval_num_tokens": 1080196.0,
+      "eval_syn_loss": 0.6980098485946655,
+      "eval_syn_runtime": 56.6342,
+      "eval_syn_samples_per_second": 24.632,
+      "eval_syn_steps_per_second": 3.09,
+      "step": 408
+    },
+    {
+      "entropy": 0.75672573694075,
+      "epoch": 1.1030674846625768,
+      "grad_norm": 0.9244198203086853,
+      "learning_rate": 0.00037356351883456205,
+      "loss": 0.7048745727539063,
+      "mean_token_accuracy": 0.783016896609104,
+      "num_tokens": 1199170.0,
+      "step": 450
+    },
+    {
+      "entropy": 0.7406797379255294,
+      "epoch": 1.2257668711656442,
+      "grad_norm": 0.6744896769523621,
+      "learning_rate": 0.00037311248151765587,
+      "loss": 0.6860730743408203,
+      "mean_token_accuracy": 0.787369327545166,
+      "num_tokens": 1334298.0,
+      "step": 500
+    },
+    {
+      "entropy": 0.7424973237514496,
+      "epoch": 1.3484662576687116,
+      "grad_norm": 0.92305588722229,
+      "learning_rate": 0.000372320629215228,
+      "loss": 0.6855724334716797,
+      "mean_token_accuracy": 0.7871134179830551,
+      "num_tokens": 1463189.0,
+      "step": 550
+    },
+    {
+      "entropy": 0.7502243828773498,
+      "epoch": 1.471165644171779,
+      "grad_norm": 0.8259047269821167,
+      "learning_rate": 0.00037118941074037944,
+      "loss": 0.6949834442138672,
+      "mean_token_accuracy": 0.7851914083957672,
+      "num_tokens": 1590310.0,
+      "step": 600
+    },
+    {
+      "entropy": 0.7233260518312454,
+      "epoch": 1.5938650306748468,
+      "grad_norm": 0.882722795009613,
+      "learning_rate": 0.00036972089582775814,
+      "loss": 0.6649005889892579,
+      "mean_token_accuracy": 0.7898879665136337,
+      "num_tokens": 1721836.0,
+      "step": 650
+    },
+    {
+      "entropy": 0.7162708967924118,
+      "epoch": 1.716564417177914,
+      "grad_norm": 0.6593027114868164,
+      "learning_rate": 0.0003679177713466678,
+      "loss": 0.6653710174560546,
+      "mean_token_accuracy": 0.7905678844451904,
+      "num_tokens": 1853507.0,
+      "step": 700
+    },
+    {
+      "entropy": 0.7150520020723343,
+      "epoch": 1.8392638036809816,
+      "grad_norm": 0.665340781211853,
+      "learning_rate": 0.0003657833363850354,
+      "loss": 0.6600938415527344,
+      "mean_token_accuracy": 0.7918649202585221,
+      "num_tokens": 1988113.0,
+      "step": 750
+    },
+    {
+      "entropy": 0.6993386310338974,
+      "epoch": 1.961963190184049,
+      "grad_norm": 0.6502187252044678,
+      "learning_rate": 0.00036332149621323294,
+      "loss": 0.6469607543945313,
+      "mean_token_accuracy": 0.7950778317451477,
+      "num_tokens": 2121314.0,
+      "step": 800
+    },
+    {
+      "epoch": 2.0,
+      "eval_entropy": 0.7315139893123082,
+      "eval_mean_token_accuracy": 0.7713504004478454,
+      "eval_not_syn_loss": 0.699893593788147,
+      "eval_not_syn_runtime": 55.4135,
+      "eval_not_syn_samples_per_second": 25.174,
+      "eval_not_syn_steps_per_second": 3.158,
+      "eval_num_tokens": 2160392.0,
+      "step": 816
+    },
+    {
+      "epoch": 2.0,
+      "eval_entropy": 0.6848451798302787,
+      "eval_mean_token_accuracy": 0.814895657471248,
+      "eval_num_tokens": 2160392.0,
+      "eval_syn_loss": 0.6415576934814453,
+      "eval_syn_runtime": 56.4735,
+      "eval_syn_samples_per_second": 24.702,
+      "eval_syn_steps_per_second": 3.099,
+      "step": 816
+    },
+    {
+      "entropy": 0.6596363644407253,
+      "epoch": 2.083435582822086,
+      "grad_norm": 0.6647700071334839,
+      "learning_rate": 0.00036053675513879666,
+      "loss": 0.599151611328125,
+      "mean_token_accuracy": 0.8071986864311527,
+      "num_tokens": 2253179.0,
+      "step": 850
+    },
+    {
+      "entropy": 0.6626429110765457,
+      "epoch": 2.2061349693251535,
+      "grad_norm": 0.7356868386268616,
+      "learning_rate": 0.00035743420826511783,
+      "loss": 0.6004195404052735,
+      "mean_token_accuracy": 0.805362731218338,
+      "num_tokens": 2379965.0,
+      "step": 900
+    },
+    {
+      "entropy": 0.6503470712900161,
+      "epoch": 2.3288343558282207,
+      "grad_norm": 1.095608115196228,
+      "learning_rate": 0.0003540195321691835,
+      "loss": 0.5951333999633789,
+      "mean_token_accuracy": 0.807324025630951,
+      "num_tokens": 2515593.0,
+      "step": 950
+    },
+    {
+      "entropy": 0.660577870607376,
+      "epoch": 2.4515337423312884,
+      "grad_norm": 0.6602179408073425,
+      "learning_rate": 0.00035029897451542325,
+      "loss": 0.6056365585327148,
+      "mean_token_accuracy": 0.8057438349723816,
+      "num_tokens": 2642790.0,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6555853113532066,
+      "epoch": 2.574233128834356,
+      "grad_norm": 0.7459151148796082,
+      "learning_rate": 0.00034627934262466606,
+      "loss": 0.5963204193115235,
+      "mean_token_accuracy": 0.8060598099231719,
+      "num_tokens": 2774396.0,
+      "step": 1050
+    },
+    {
+      "entropy": 0.6505853188037872,
+      "epoch": 2.6969325153374233,
+      "grad_norm": 0.6359792947769165,
+      "learning_rate": 0.00034196799101912103,
+      "loss": 0.5993065643310547,
+      "mean_token_accuracy": 0.806280387043953,
+      "num_tokens": 2912102.0,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6509970253705979,
+      "epoch": 2.819631901840491,
+      "grad_norm": 0.8086174726486206,
+      "learning_rate": 0.0003373728079661707,
+      "loss": 0.6001268005371094,
+      "mean_token_accuracy": 0.806866540312767,
+      "num_tokens": 3042987.0,
+      "step": 1150
+    },
+    {
+      "entropy": 0.6404844307899475,
+      "epoch": 2.942331288343558,
+      "grad_norm": 0.7038407325744629,
+      "learning_rate": 0.0003325022010455975,
+      "loss": 0.5870526123046875,
+      "mean_token_accuracy": 0.8104586935043335,
+      "num_tokens": 3178044.0,
+      "step": 1200
+    },
+    {
+      "epoch": 3.0,
+      "eval_entropy": 0.6645996727262224,
+      "eval_mean_token_accuracy": 0.798654237134116,
+      "eval_not_syn_loss": 0.6586771011352539,
+      "eval_not_syn_runtime": 55.1088,
+      "eval_not_syn_samples_per_second": 25.314,
+      "eval_not_syn_steps_per_second": 3.176,
+      "eval_num_tokens": 3240588.0,
+      "step": 1224
+    },
+    {
+      "epoch": 3.0,
+      "eval_entropy": 0.6233607809884207,
+      "eval_mean_token_accuracy": 0.7977249867575509,
+      "eval_num_tokens": 3240588.0,
+      "eval_syn_loss": 0.6288298964500427,
+      "eval_syn_runtime": 56.6081,
+      "eval_syn_samples_per_second": 24.643,
+      "eval_syn_steps_per_second": 3.091,
+      "step": 1224
+    },
+    {
+      "entropy": 0.6081961104364106,
+      "epoch": 3.063803680981595,
+      "grad_norm": 0.6161203980445862,
+      "learning_rate": 0.0003273650817666497,
+      "loss": 0.5481716537475586,
+      "mean_token_accuracy": 0.8175542035488167,
+      "num_tokens": 3312797.0,
+      "step": 1250
+    },
+    {
+      "entropy": 0.5877064895629883,
+      "epoch": 3.1865030674846624,
+      "grad_norm": 0.6909636855125427,
+      "learning_rate": 0.00032197084926309234,
+      "loss": 0.5224150466918945,
+      "mean_token_accuracy": 0.8250984263420105,
+      "num_tokens": 3441629.0,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5952656552195549,
+      "epoch": 3.30920245398773,
+      "grad_norm": 0.8147026300430298,
+      "learning_rate": 0.0003163293730960758,
+      "loss": 0.5349474334716797,
+      "mean_token_accuracy": 0.8213064837455749,
+      "num_tokens": 3576093.0,
+      "step": 1350
+    },
+    {
+      "entropy": 0.5937123569846153,
+      "epoch": 3.4319018404907977,
+      "grad_norm": 0.8690970540046692,
+      "learning_rate": 0.0003104509751962869,
+      "loss": 0.5310481262207031,
+      "mean_token_accuracy": 0.8229135835170746,
+      "num_tokens": 3705912.0,
+      "step": 1400
+    },
+    {
+      "entropy": 0.6067730465531349,
+      "epoch": 3.554601226993865,
+      "grad_norm": 0.7433330416679382,
+      "learning_rate": 0.0003043464109784207,
+      "loss": 0.5450045394897461,
+      "mean_token_accuracy": 0.8205251175165177,
+      "num_tokens": 3832376.0,
+      "step": 1450
+    },
+    {
+      "entropy": 0.5926914718747139,
+      "epoch": 3.6773006134969326,
+      "grad_norm": 0.7265587449073792,
+      "learning_rate": 0.00029802684966252974,
+      "loss": 0.5330168914794922,
+      "mean_token_accuracy": 0.821294127702713,
+      "num_tokens": 3967273.0,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5881199997663498,
+      "epoch": 3.8,
+      "grad_norm": 0.7681817412376404,
+      "learning_rate": 0.00029150385383825187,
+      "loss": 0.5288929367065429,
+      "mean_token_accuracy": 0.8234161508083343,
+      "num_tokens": 4102434.0,
+      "step": 1550
+    },
+    {
+      "entropy": 0.6008686199784279,
+      "epoch": 3.9226993865030675,
+      "grad_norm": 0.9384056329727173,
+      "learning_rate": 0.0002847893583093113,
+      "loss": 0.5437511444091797,
+      "mean_token_accuracy": 0.8201366275548935,
+      "num_tokens": 4234037.0,
+      "step": 1600
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.6218148783275059,
+      "eval_mean_token_accuracy": 0.783796865599496,
+      "eval_not_syn_loss": 0.6558170318603516,
+      "eval_not_syn_runtime": 55.429,
+      "eval_not_syn_samples_per_second": 25.167,
+      "eval_not_syn_steps_per_second": 3.157,
+      "eval_num_tokens": 4320784.0,
+      "step": 1632
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5833754854542869,
+      "eval_mean_token_accuracy": 0.8278223422595433,
+      "eval_num_tokens": 4320784.0,
+      "eval_syn_loss": 0.604770302772522,
+      "eval_syn_runtime": 56.9593,
+      "eval_syn_samples_per_second": 24.491,
+      "eval_syn_steps_per_second": 3.072,
+      "step": 1632
+    },
+    {
+      "entropy": 0.559690699733869,
+      "epoch": 4.044171779141104,
+      "grad_norm": 0.65739905834198,
+      "learning_rate": 0.0002778956482569963,
+      "loss": 0.49702674865722657,
+      "mean_token_accuracy": 0.8317034166268628,
+      "num_tokens": 4368268.0,
+      "step": 1650
+    },
+    {
+      "entropy": 0.5401161375641823,
+      "epoch": 4.166871165644172,
+      "grad_norm": 1.5176353454589844,
+      "learning_rate": 0.0002708353367625691,
+      "loss": 0.4598746109008789,
+      "mean_token_accuracy": 0.8412033343315124,
+      "num_tokens": 4494525.0,
+      "step": 1700
+    },
+    {
+      "entropy": 0.5299916860461235,
+      "epoch": 4.289570552147239,
+      "grad_norm": 0.685965359210968,
+      "learning_rate": 0.0002636213417297333,
+      "loss": 0.45663665771484374,
+      "mean_token_accuracy": 0.8412413811683654,
+      "num_tokens": 4631985.0,
+      "step": 1750
+    },
+    {
+      "entropy": 0.5352088099718094,
+      "epoch": 4.412269938650307,
+      "grad_norm": 0.8471158742904663,
+      "learning_rate": 0.0002562668622493824,
+      "loss": 0.4625707626342773,
+      "mean_token_accuracy": 0.8400991547107697,
+      "num_tokens": 4766323.0,
+      "step": 1800
+    },
+    {
+      "entropy": 0.5425067806243896,
+      "epoch": 4.534969325153375,
+      "grad_norm": 0.8595437407493591,
+      "learning_rate": 0.00024878535444987474,
+      "loss": 0.47124954223632814,
+      "mean_token_accuracy": 0.8376108628511428,
+      "num_tokens": 4895606.0,
+      "step": 1850
+    },
+    {
+      "entropy": 0.5419312778115273,
+      "epoch": 4.6576687116564415,
+      "grad_norm": 0.8436959981918335,
+      "learning_rate": 0.00024119050687701818,
+      "loss": 0.4708108901977539,
+      "mean_token_accuracy": 0.8390797019004822,
+      "num_tokens": 5025912.0,
+      "step": 1900
+    },
+    {
+      "entropy": 0.5392271488904953,
+      "epoch": 4.780368098159509,
+      "grad_norm": 0.8691944479942322,
+      "learning_rate": 0.00023349621544881404,
+      "loss": 0.46682029724121094,
+      "mean_token_accuracy": 0.8385172665119172,
+      "num_tokens": 5162972.0,
+      "step": 1950
+    },
+    {
+      "entropy": 0.5332251599431038,
+      "epoch": 4.903067484662577,
+      "grad_norm": 0.6814598441123962,
+      "learning_rate": 0.00022571655803078078,
+      "loss": 0.4687493896484375,
+      "mean_token_accuracy": 0.8388861840963364,
+      "num_tokens": 5297935.0,
+      "step": 2000
+    },
+    {
+      "epoch": 5.0,
+      "eval_entropy": 0.5847777877535139,
+      "eval_mean_token_accuracy": 0.7782382583618164,
+      "eval_not_syn_loss": 0.6678160429000854,
+      "eval_not_syn_runtime": 55.1467,
+      "eval_not_syn_samples_per_second": 25.296,
+      "eval_not_syn_steps_per_second": 3.173,
+      "eval_num_tokens": 5400980.0,
+      "step": 2040
+    },
+    {
+      "epoch": 5.0,
+      "eval_entropy": 0.5505962218557086,
+      "eval_mean_token_accuracy": 0.8308270008223397,
+      "eval_num_tokens": 5400980.0,
+      "eval_syn_loss": 0.606867253780365,
+      "eval_syn_runtime": 56.9768,
+      "eval_syn_samples_per_second": 24.484,
+      "eval_syn_steps_per_second": 3.071,
+      "step": 2040
+    },
+    {
+      "entropy": 0.533086326688227,
+      "epoch": 5.024539877300613,
+      "grad_norm": 0.9453939199447632,
+      "learning_rate": 0.00021786576867837776,
+      "loss": 0.45802555084228513,
+      "mean_token_accuracy": 0.8429379854539428,
+      "num_tokens": 5427362.0,
+      "step": 2050
+    },
+    {
+      "entropy": 0.46721911430358887,
+      "epoch": 5.147239263803681,
+      "grad_norm": 0.9237910509109497,
+      "learning_rate": 0.00020995821159365647,
+      "loss": 0.37914146423339845,
+      "mean_token_accuracy": 0.8648965793848038,
+      "num_tokens": 5564186.0,
+      "step": 2100
+    },
+    {
+      "entropy": 0.4682381117343903,
+      "epoch": 5.269938650306749,
+      "grad_norm": 0.7424130439758301,
+      "learning_rate": 0.0002020083548437879,
+      "loss": 0.3859384536743164,
+      "mean_token_accuracy": 0.8622810471057892,
+      "num_tokens": 5694728.0,
+      "step": 2150
+    },
+    {
+      "entropy": 0.464998793900013,
+      "epoch": 5.392638036809816,
+      "grad_norm": 0.9224902987480164,
+      "learning_rate": 0.0001940307438895544,
+      "loss": 0.3839593505859375,
+      "mean_token_accuracy": 0.8627353543043137,
+      "num_tokens": 5832581.0,
+      "step": 2200
+    },
+    {
+      "entropy": 0.4753906837105751,
+      "epoch": 5.515337423312883,
+      "grad_norm": 0.8931427001953125,
+      "learning_rate": 0.0001860399749722367,
+      "loss": 0.393004264831543,
+      "mean_token_accuracy": 0.8586806547641754,
+      "num_tokens": 5962417.0,
+      "step": 2250
+    },
+    {
+      "entropy": 0.4671635267138481,
+      "epoch": 5.638036809815951,
+      "grad_norm": 0.9559172987937927,
+      "learning_rate": 0.00017805066840759193,
+      "loss": 0.38852767944335936,
+      "mean_token_accuracy": 0.8615026396512985,
+      "num_tokens": 6092338.0,
+      "step": 2300
+    },
+    {
+      "entropy": 0.4746100226044655,
+      "epoch": 5.7607361963190185,
+      "grad_norm": 0.9070253968238831,
+      "learning_rate": 0.00017007744183578254,
+      "loss": 0.3960166168212891,
+      "mean_token_accuracy": 0.8591635781526565,
+      "num_tokens": 6226327.0,
+      "step": 2350
+    },
+    {
+      "entropy": 0.4698092398047447,
+      "epoch": 5.883435582822086,
+      "grad_norm": 0.8354232907295227,
+      "learning_rate": 0.0001621348834762011,
+      "loss": 0.39321571350097656,
+      "mean_token_accuracy": 0.8603039687871933,
+      "num_tokens": 6359083.0,
+      "step": 2400
+    },
+    {
+      "epoch": 6.0,
+      "eval_entropy": 0.536405998979296,
+      "eval_mean_token_accuracy": 0.790239520413535,
+      "eval_not_syn_loss": 0.6884409785270691,
+      "eval_not_syn_runtime": 54.9927,
+      "eval_not_syn_samples_per_second": 25.367,
+      "eval_not_syn_steps_per_second": 3.182,
+      "eval_num_tokens": 6481176.0,
+      "step": 2448
+    },
+    {
+      "epoch": 6.0,
+      "eval_entropy": 0.5039791010107313,
+      "eval_mean_token_accuracy": 0.8163431504794529,
+      "eval_num_tokens": 6481176.0,
+      "eval_syn_loss": 0.6377155184745789,
+      "eval_syn_runtime": 57.196,
+      "eval_syn_samples_per_second": 24.39,
+      "eval_syn_steps_per_second": 3.06,
+      "step": 2448
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 4080,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.2906721035504346e+17,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2856/README.md b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2856/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..7ac50e2706e5b4eae8f28e204601afb241c49e55
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2856/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-2B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-2B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2856/adapter_config.json b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2856/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b9596096c71017d49955136a4e23d9f97138d305
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2856/adapter_config.json
@@ -0,0 +1,48 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.08602048083239339,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "q_proj",
+    "k_proj",
+    "gate_proj",
+    "up_proj",
+    "o_proj",
+    "down_proj",
+    "v_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2856/chat_template.jinja b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2856/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..0ef09f214eaa6d9bca297988afc1454b5827b2c7
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2856/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is true %}
+        {{- '<think>\n' }}
+    {%- else %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2856/tokenizer_config.json b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2856/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2856/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2856/trainer_state.json b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2856/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..18f1c5dc055d9ae52a37ef1eca52b5c209d3ca28
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-2856/trainer_state.json
@@ -0,0 +1,758 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 7.0,
+  "eval_steps": 500,
+  "global_step": 2856,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.2408065378665922,
+      "epoch": 0.12269938650306748,
+      "grad_norm": 1.6718658208847046,
+      "learning_rate": 4.487804970912036e-05,
+      "loss": 2.129624786376953,
+      "mean_token_accuracy": 0.5743644836544991,
+      "num_tokens": 134329.0,
+      "step": 50
+    },
+    {
+      "entropy": 1.1078504550457,
+      "epoch": 0.24539877300613497,
+      "grad_norm": 2.4091813564300537,
+      "learning_rate": 9.067197798373296e-05,
+      "loss": 1.0346390533447265,
+      "mean_token_accuracy": 0.7085381114482879,
+      "num_tokens": 262407.0,
+      "step": 100
+    },
+    {
+      "entropy": 0.9421598988771439,
+      "epoch": 0.36809815950920244,
+      "grad_norm": 1.3913646936416626,
+      "learning_rate": 0.00013646590625834558,
+      "loss": 0.8716551208496094,
+      "mean_token_accuracy": 0.7449040985107422,
+      "num_tokens": 401136.0,
+      "step": 150
+    },
+    {
+      "entropy": 0.8723550146818161,
+      "epoch": 0.49079754601226994,
+      "grad_norm": 1.1664468050003052,
+      "learning_rate": 0.0001822598345329582,
+      "loss": 0.8095977783203125,
+      "mean_token_accuracy": 0.7585240352153778,
+      "num_tokens": 531647.0,
+      "step": 200
+    },
+    {
+      "entropy": 0.8466519421339035,
+      "epoch": 0.6134969325153374,
+      "grad_norm": 1.0145421028137207,
+      "learning_rate": 0.00022805376280757083,
+      "loss": 0.7830857849121093,
+      "mean_token_accuracy": 0.7661705195903779,
+      "num_tokens": 666127.0,
+      "step": 250
+    },
+    {
+      "entropy": 0.8180295366048813,
+      "epoch": 0.7361963190184049,
+      "grad_norm": 0.9731705784797668,
+      "learning_rate": 0.00027384769108218336,
+      "loss": 0.7573603820800782,
+      "mean_token_accuracy": 0.7704453605413437,
+      "num_tokens": 799292.0,
+      "step": 300
+    },
+    {
+      "entropy": 0.7990227991342544,
+      "epoch": 0.8588957055214724,
+      "grad_norm": 1.0642374753952026,
+      "learning_rate": 0.000319641619356796,
+      "loss": 0.73709228515625,
+      "mean_token_accuracy": 0.7752733880281448,
+      "num_tokens": 930614.0,
+      "step": 350
+    },
+    {
+      "entropy": 0.7899087685346603,
+      "epoch": 0.9815950920245399,
+      "grad_norm": 0.880613386631012,
+      "learning_rate": 0.0003654355476314086,
+      "loss": 0.7347319793701171,
+      "mean_token_accuracy": 0.7766731631755829,
+      "num_tokens": 1061654.0,
+      "step": 400
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.83522472347532,
+      "eval_mean_token_accuracy": 0.7484098076820374,
+      "eval_not_syn_loss": 0.7872886061668396,
+      "eval_not_syn_runtime": 54.8697,
+      "eval_not_syn_samples_per_second": 25.424,
+      "eval_not_syn_steps_per_second": 3.189,
+      "eval_num_tokens": 1080196.0,
+      "step": 408
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.7784442271505083,
+      "eval_mean_token_accuracy": 0.8021791185651507,
+      "eval_num_tokens": 1080196.0,
+      "eval_syn_loss": 0.6980098485946655,
+      "eval_syn_runtime": 56.6342,
+      "eval_syn_samples_per_second": 24.632,
+      "eval_syn_steps_per_second": 3.09,
+      "step": 408
+    },
+    {
+      "entropy": 0.75672573694075,
+      "epoch": 1.1030674846625768,
+      "grad_norm": 0.9244198203086853,
+      "learning_rate": 0.00037356351883456205,
+      "loss": 0.7048745727539063,
+      "mean_token_accuracy": 0.783016896609104,
+      "num_tokens": 1199170.0,
+      "step": 450
+    },
+    {
+      "entropy": 0.7406797379255294,
+      "epoch": 1.2257668711656442,
+      "grad_norm": 0.6744896769523621,
+      "learning_rate": 0.00037311248151765587,
+      "loss": 0.6860730743408203,
+      "mean_token_accuracy": 0.787369327545166,
+      "num_tokens": 1334298.0,
+      "step": 500
+    },
+    {
+      "entropy": 0.7424973237514496,
+      "epoch": 1.3484662576687116,
+      "grad_norm": 0.92305588722229,
+      "learning_rate": 0.000372320629215228,
+      "loss": 0.6855724334716797,
+      "mean_token_accuracy": 0.7871134179830551,
+      "num_tokens": 1463189.0,
+      "step": 550
+    },
+    {
+      "entropy": 0.7502243828773498,
+      "epoch": 1.471165644171779,
+      "grad_norm": 0.8259047269821167,
+      "learning_rate": 0.00037118941074037944,
+      "loss": 0.6949834442138672,
+      "mean_token_accuracy": 0.7851914083957672,
+      "num_tokens": 1590310.0,
+      "step": 600
+    },
+    {
+      "entropy": 0.7233260518312454,
+      "epoch": 1.5938650306748468,
+      "grad_norm": 0.882722795009613,
+      "learning_rate": 0.00036972089582775814,
+      "loss": 0.6649005889892579,
+      "mean_token_accuracy": 0.7898879665136337,
+      "num_tokens": 1721836.0,
+      "step": 650
+    },
+    {
+      "entropy": 0.7162708967924118,
+      "epoch": 1.716564417177914,
+      "grad_norm": 0.6593027114868164,
+      "learning_rate": 0.0003679177713466678,
+      "loss": 0.6653710174560546,
+      "mean_token_accuracy": 0.7905678844451904,
+      "num_tokens": 1853507.0,
+      "step": 700
+    },
+    {
+      "entropy": 0.7150520020723343,
+      "epoch": 1.8392638036809816,
+      "grad_norm": 0.665340781211853,
+      "learning_rate": 0.0003657833363850354,
+      "loss": 0.6600938415527344,
+      "mean_token_accuracy": 0.7918649202585221,
+      "num_tokens": 1988113.0,
+      "step": 750
+    },
+    {
+      "entropy": 0.6993386310338974,
+      "epoch": 1.961963190184049,
+      "grad_norm": 0.6502187252044678,
+      "learning_rate": 0.00036332149621323294,
+      "loss": 0.6469607543945313,
+      "mean_token_accuracy": 0.7950778317451477,
+      "num_tokens": 2121314.0,
+      "step": 800
+    },
+    {
+      "epoch": 2.0,
+      "eval_entropy": 0.7315139893123082,
+      "eval_mean_token_accuracy": 0.7713504004478454,
+      "eval_not_syn_loss": 0.699893593788147,
+      "eval_not_syn_runtime": 55.4135,
+      "eval_not_syn_samples_per_second": 25.174,
+      "eval_not_syn_steps_per_second": 3.158,
+      "eval_num_tokens": 2160392.0,
+      "step": 816
+    },
+    {
+      "epoch": 2.0,
+      "eval_entropy": 0.6848451798302787,
+      "eval_mean_token_accuracy": 0.814895657471248,
+      "eval_num_tokens": 2160392.0,
+      "eval_syn_loss": 0.6415576934814453,
+      "eval_syn_runtime": 56.4735,
+      "eval_syn_samples_per_second": 24.702,
+      "eval_syn_steps_per_second": 3.099,
+      "step": 816
+    },
+    {
+      "entropy": 0.6596363644407253,
+      "epoch": 2.083435582822086,
+      "grad_norm": 0.6647700071334839,
+      "learning_rate": 0.00036053675513879666,
+      "loss": 0.599151611328125,
+      "mean_token_accuracy": 0.8071986864311527,
+      "num_tokens": 2253179.0,
+      "step": 850
+    },
+    {
+      "entropy": 0.6626429110765457,
+      "epoch": 2.2061349693251535,
+      "grad_norm": 0.7356868386268616,
+      "learning_rate": 0.00035743420826511783,
+      "loss": 0.6004195404052735,
+      "mean_token_accuracy": 0.805362731218338,
+      "num_tokens": 2379965.0,
+      "step": 900
+    },
+    {
+      "entropy": 0.6503470712900161,
+      "epoch": 2.3288343558282207,
+      "grad_norm": 1.095608115196228,
+      "learning_rate": 0.0003540195321691835,
+      "loss": 0.5951333999633789,
+      "mean_token_accuracy": 0.807324025630951,
+      "num_tokens": 2515593.0,
+      "step": 950
+    },
+    {
+      "entropy": 0.660577870607376,
+      "epoch": 2.4515337423312884,
+      "grad_norm": 0.6602179408073425,
+      "learning_rate": 0.00035029897451542325,
+      "loss": 0.6056365585327148,
+      "mean_token_accuracy": 0.8057438349723816,
+      "num_tokens": 2642790.0,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6555853113532066,
+      "epoch": 2.574233128834356,
+      "grad_norm": 0.7459151148796082,
+      "learning_rate": 0.00034627934262466606,
+      "loss": 0.5963204193115235,
+      "mean_token_accuracy": 0.8060598099231719,
+      "num_tokens": 2774396.0,
+      "step": 1050
+    },
+    {
+      "entropy": 0.6505853188037872,
+      "epoch": 2.6969325153374233,
+      "grad_norm": 0.6359792947769165,
+      "learning_rate": 0.00034196799101912103,
+      "loss": 0.5993065643310547,
+      "mean_token_accuracy": 0.806280387043953,
+      "num_tokens": 2912102.0,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6509970253705979,
+      "epoch": 2.819631901840491,
+      "grad_norm": 0.8086174726486206,
+      "learning_rate": 0.0003373728079661707,
+      "loss": 0.6001268005371094,
+      "mean_token_accuracy": 0.806866540312767,
+      "num_tokens": 3042987.0,
+      "step": 1150
+    },
+    {
+      "entropy": 0.6404844307899475,
+      "epoch": 2.942331288343558,
+      "grad_norm": 0.7038407325744629,
+      "learning_rate": 0.0003325022010455975,
+      "loss": 0.5870526123046875,
+      "mean_token_accuracy": 0.8104586935043335,
+      "num_tokens": 3178044.0,
+      "step": 1200
+    },
+    {
+      "epoch": 3.0,
+      "eval_entropy": 0.6645996727262224,
+      "eval_mean_token_accuracy": 0.798654237134116,
+      "eval_not_syn_loss": 0.6586771011352539,
+      "eval_not_syn_runtime": 55.1088,
+      "eval_not_syn_samples_per_second": 25.314,
+      "eval_not_syn_steps_per_second": 3.176,
+      "eval_num_tokens": 3240588.0,
+      "step": 1224
+    },
+    {
+      "epoch": 3.0,
+      "eval_entropy": 0.6233607809884207,
+      "eval_mean_token_accuracy": 0.7977249867575509,
+      "eval_num_tokens": 3240588.0,
+      "eval_syn_loss": 0.6288298964500427,
+      "eval_syn_runtime": 56.6081,
+      "eval_syn_samples_per_second": 24.643,
+      "eval_syn_steps_per_second": 3.091,
+      "step": 1224
+    },
+    {
+      "entropy": 0.6081961104364106,
+      "epoch": 3.063803680981595,
+      "grad_norm": 0.6161203980445862,
+      "learning_rate": 0.0003273650817666497,
+      "loss": 0.5481716537475586,
+      "mean_token_accuracy": 0.8175542035488167,
+      "num_tokens": 3312797.0,
+      "step": 1250
+    },
+    {
+      "entropy": 0.5877064895629883,
+      "epoch": 3.1865030674846624,
+      "grad_norm": 0.6909636855125427,
+      "learning_rate": 0.00032197084926309234,
+      "loss": 0.5224150466918945,
+      "mean_token_accuracy": 0.8250984263420105,
+      "num_tokens": 3441629.0,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5952656552195549,
+      "epoch": 3.30920245398773,
+      "grad_norm": 0.8147026300430298,
+      "learning_rate": 0.0003163293730960758,
+      "loss": 0.5349474334716797,
+      "mean_token_accuracy": 0.8213064837455749,
+      "num_tokens": 3576093.0,
+      "step": 1350
+    },
+    {
+      "entropy": 0.5937123569846153,
+      "epoch": 3.4319018404907977,
+      "grad_norm": 0.8690970540046692,
+      "learning_rate": 0.0003104509751962869,
+      "loss": 0.5310481262207031,
+      "mean_token_accuracy": 0.8229135835170746,
+      "num_tokens": 3705912.0,
+      "step": 1400
+    },
+    {
+      "entropy": 0.6067730465531349,
+      "epoch": 3.554601226993865,
+      "grad_norm": 0.7433330416679382,
+      "learning_rate": 0.0003043464109784207,
+      "loss": 0.5450045394897461,
+      "mean_token_accuracy": 0.8205251175165177,
+      "num_tokens": 3832376.0,
+      "step": 1450
+    },
+    {
+      "entropy": 0.5926914718747139,
+      "epoch": 3.6773006134969326,
+      "grad_norm": 0.7265587449073792,
+      "learning_rate": 0.00029802684966252974,
+      "loss": 0.5330168914794922,
+      "mean_token_accuracy": 0.821294127702713,
+      "num_tokens": 3967273.0,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5881199997663498,
+      "epoch": 3.8,
+      "grad_norm": 0.7681817412376404,
+      "learning_rate": 0.00029150385383825187,
+      "loss": 0.5288929367065429,
+      "mean_token_accuracy": 0.8234161508083343,
+      "num_tokens": 4102434.0,
+      "step": 1550
+    },
+    {
+      "entropy": 0.6008686199784279,
+      "epoch": 3.9226993865030675,
+      "grad_norm": 0.9384056329727173,
+      "learning_rate": 0.0002847893583093113,
+      "loss": 0.5437511444091797,
+      "mean_token_accuracy": 0.8201366275548935,
+      "num_tokens": 4234037.0,
+      "step": 1600
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.6218148783275059,
+      "eval_mean_token_accuracy": 0.783796865599496,
+      "eval_not_syn_loss": 0.6558170318603516,
+      "eval_not_syn_runtime": 55.429,
+      "eval_not_syn_samples_per_second": 25.167,
+      "eval_not_syn_steps_per_second": 3.157,
+      "eval_num_tokens": 4320784.0,
+      "step": 1632
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5833754854542869,
+      "eval_mean_token_accuracy": 0.8278223422595433,
+      "eval_num_tokens": 4320784.0,
+      "eval_syn_loss": 0.604770302772522,
+      "eval_syn_runtime": 56.9593,
+      "eval_syn_samples_per_second": 24.491,
+      "eval_syn_steps_per_second": 3.072,
+      "step": 1632
+    },
+    {
+      "entropy": 0.559690699733869,
+      "epoch": 4.044171779141104,
+      "grad_norm": 0.65739905834198,
+      "learning_rate": 0.0002778956482569963,
+      "loss": 0.49702674865722657,
+      "mean_token_accuracy": 0.8317034166268628,
+      "num_tokens": 4368268.0,
+      "step": 1650
+    },
+    {
+      "entropy": 0.5401161375641823,
+      "epoch": 4.166871165644172,
+      "grad_norm": 1.5176353454589844,
+      "learning_rate": 0.0002708353367625691,
+      "loss": 0.4598746109008789,
+      "mean_token_accuracy": 0.8412033343315124,
+      "num_tokens": 4494525.0,
+      "step": 1700
+    },
+    {
+      "entropy": 0.5299916860461235,
+      "epoch": 4.289570552147239,
+      "grad_norm": 0.685965359210968,
+      "learning_rate": 0.0002636213417297333,
+      "loss": 0.45663665771484374,
+      "mean_token_accuracy": 0.8412413811683654,
+      "num_tokens": 4631985.0,
+      "step": 1750
+    },
+    {
+      "entropy": 0.5352088099718094,
+      "epoch": 4.412269938650307,
+      "grad_norm": 0.8471158742904663,
+      "learning_rate": 0.0002562668622493824,
+      "loss": 0.4625707626342773,
+      "mean_token_accuracy": 0.8400991547107697,
+      "num_tokens": 4766323.0,
+      "step": 1800
+    },
+    {
+      "entropy": 0.5425067806243896,
+      "epoch": 4.534969325153375,
+      "grad_norm": 0.8595437407493591,
+      "learning_rate": 0.00024878535444987474,
+      "loss": 0.47124954223632814,
+      "mean_token_accuracy": 0.8376108628511428,
+      "num_tokens": 4895606.0,
+      "step": 1850
+    },
+    {
+      "entropy": 0.5419312778115273,
+      "epoch": 4.6576687116564415,
+      "grad_norm": 0.8436959981918335,
+      "learning_rate": 0.00024119050687701818,
+      "loss": 0.4708108901977539,
+      "mean_token_accuracy": 0.8390797019004822,
+      "num_tokens": 5025912.0,
+      "step": 1900
+    },
+    {
+      "entropy": 0.5392271488904953,
+      "epoch": 4.780368098159509,
+      "grad_norm": 0.8691944479942322,
+      "learning_rate": 0.00023349621544881404,
+      "loss": 0.46682029724121094,
+      "mean_token_accuracy": 0.8385172665119172,
+      "num_tokens": 5162972.0,
+      "step": 1950
+    },
+    {
+      "entropy": 0.5332251599431038,
+      "epoch": 4.903067484662577,
+      "grad_norm": 0.6814598441123962,
+      "learning_rate": 0.00022571655803078078,
+      "loss": 0.4687493896484375,
+      "mean_token_accuracy": 0.8388861840963364,
+      "num_tokens": 5297935.0,
+      "step": 2000
+    },
+    {
+      "epoch": 5.0,
+      "eval_entropy": 0.5847777877535139,
+      "eval_mean_token_accuracy": 0.7782382583618164,
+      "eval_not_syn_loss": 0.6678160429000854,
+      "eval_not_syn_runtime": 55.1467,
+      "eval_not_syn_samples_per_second": 25.296,
+      "eval_not_syn_steps_per_second": 3.173,
+      "eval_num_tokens": 5400980.0,
+      "step": 2040
+    },
+    {
+      "epoch": 5.0,
+      "eval_entropy": 0.5505962218557086,
+      "eval_mean_token_accuracy": 0.8308270008223397,
+      "eval_num_tokens": 5400980.0,
+      "eval_syn_loss": 0.606867253780365,
+      "eval_syn_runtime": 56.9768,
+      "eval_syn_samples_per_second": 24.484,
+      "eval_syn_steps_per_second": 3.071,
+      "step": 2040
+    },
+    {
+      "entropy": 0.533086326688227,
+      "epoch": 5.024539877300613,
+      "grad_norm": 0.9453939199447632,
+      "learning_rate": 0.00021786576867837776,
+      "loss": 0.45802555084228513,
+      "mean_token_accuracy": 0.8429379854539428,
+      "num_tokens": 5427362.0,
+      "step": 2050
+    },
+    {
+      "entropy": 0.46721911430358887,
+      "epoch": 5.147239263803681,
+      "grad_norm": 0.9237910509109497,
+      "learning_rate": 0.00020995821159365647,
+      "loss": 0.37914146423339845,
+      "mean_token_accuracy": 0.8648965793848038,
+      "num_tokens": 5564186.0,
+      "step": 2100
+    },
+    {
+      "entropy": 0.4682381117343903,
+      "epoch": 5.269938650306749,
+      "grad_norm": 0.7424130439758301,
+      "learning_rate": 0.0002020083548437879,
+      "loss": 0.3859384536743164,
+      "mean_token_accuracy": 0.8622810471057892,
+      "num_tokens": 5694728.0,
+      "step": 2150
+    },
+    {
+      "entropy": 0.464998793900013,
+      "epoch": 5.392638036809816,
+      "grad_norm": 0.9224902987480164,
+      "learning_rate": 0.0001940307438895544,
+      "loss": 0.3839593505859375,
+      "mean_token_accuracy": 0.8627353543043137,
+      "num_tokens": 5832581.0,
+      "step": 2200
+    },
+    {
+      "entropy": 0.4753906837105751,
+      "epoch": 5.515337423312883,
+      "grad_norm": 0.8931427001953125,
+      "learning_rate": 0.0001860399749722367,
+      "loss": 0.393004264831543,
+      "mean_token_accuracy": 0.8586806547641754,
+      "num_tokens": 5962417.0,
+      "step": 2250
+    },
+    {
+      "entropy": 0.4671635267138481,
+      "epoch": 5.638036809815951,
+      "grad_norm": 0.9559172987937927,
+      "learning_rate": 0.00017805066840759193,
+      "loss": 0.38852767944335936,
+      "mean_token_accuracy": 0.8615026396512985,
+      "num_tokens": 6092338.0,
+      "step": 2300
+    },
+    {
+      "entropy": 0.4746100226044655,
+      "epoch": 5.7607361963190185,
+      "grad_norm": 0.9070253968238831,
+      "learning_rate": 0.00017007744183578254,
+      "loss": 0.3960166168212891,
+      "mean_token_accuracy": 0.8591635781526565,
+      "num_tokens": 6226327.0,
+      "step": 2350
+    },
+    {
+      "entropy": 0.4698092398047447,
+      "epoch": 5.883435582822086,
+      "grad_norm": 0.8354232907295227,
+      "learning_rate": 0.0001621348834762011,
+      "loss": 0.39321571350097656,
+      "mean_token_accuracy": 0.8603039687871933,
+      "num_tokens": 6359083.0,
+      "step": 2400
+    },
+    {
+      "epoch": 6.0,
+      "eval_entropy": 0.536405998979296,
+      "eval_mean_token_accuracy": 0.790239520413535,
+      "eval_not_syn_loss": 0.6884409785270691,
+      "eval_not_syn_runtime": 54.9927,
+      "eval_not_syn_samples_per_second": 25.367,
+      "eval_not_syn_steps_per_second": 3.182,
+      "eval_num_tokens": 6481176.0,
+      "step": 2448
+    },
+    {
+      "epoch": 6.0,
+      "eval_entropy": 0.5039791010107313,
+      "eval_mean_token_accuracy": 0.8163431504794529,
+      "eval_num_tokens": 6481176.0,
+      "eval_syn_loss": 0.6377155184745789,
+      "eval_syn_runtime": 57.196,
+      "eval_syn_samples_per_second": 24.39,
+      "eval_syn_steps_per_second": 3.06,
+      "step": 2448
+    },
+    {
+      "entropy": 0.4671963176341972,
+      "epoch": 6.004907975460123,
+      "grad_norm": 0.8445424437522888,
+      "learning_rate": 0.00015423752543612516,
+      "loss": 0.3872825241088867,
+      "mean_token_accuracy": 0.8618519083418027,
+      "num_tokens": 6486549.0,
+      "step": 2450
+    },
+    {
+      "entropy": 0.3890652832388878,
+      "epoch": 6.12760736196319,
+      "grad_norm": 0.725812554359436,
+      "learning_rate": 0.00014639981712203683,
+      "loss": 0.2923568153381348,
+      "mean_token_accuracy": 0.8928290349245072,
+      "num_tokens": 6622384.0,
+      "step": 2500
+    },
+    {
+      "entropy": 0.39649384289979933,
+      "epoch": 6.250306748466258,
+      "grad_norm": 0.9143152236938477,
+      "learning_rate": 0.000138636098802257,
+      "loss": 0.303057861328125,
+      "mean_token_accuracy": 0.8888710218667984,
+      "num_tokens": 6758637.0,
+      "step": 2550
+    },
+    {
+      "entropy": 0.4020465177297592,
+      "epoch": 6.373006134969325,
+      "grad_norm": 0.9407342076301575,
+      "learning_rate": 0.00013096057536926304,
+      "loss": 0.3073894691467285,
+      "mean_token_accuracy": 0.8894575881958008,
+      "num_tokens": 6888533.0,
+      "step": 2600
+    },
+    {
+      "entropy": 0.397317076921463,
+      "epoch": 6.495705521472392,
+      "grad_norm": 0.8517777323722839,
+      "learning_rate": 0.00012338729034969828,
+      "loss": 0.30508209228515626,
+      "mean_token_accuracy": 0.8886856293678284,
+      "num_tokens": 7023729.0,
+      "step": 2650
+    },
+    {
+      "entropy": 0.3915065750479698,
+      "epoch": 6.61840490797546,
+      "grad_norm": 0.862319827079773,
+      "learning_rate": 0.00011593010020962378,
+      "loss": 0.30237462997436526,
+      "mean_token_accuracy": 0.8897094148397445,
+      "num_tokens": 7153821.0,
+      "step": 2700
+    },
+    {
+      "entropy": 0.3977497002482414,
+      "epoch": 6.741104294478528,
+      "grad_norm": 0.8821370601654053,
+      "learning_rate": 0.00010860264900202563,
+      "loss": 0.3067088508605957,
+      "mean_token_accuracy": 0.8873285222053527,
+      "num_tokens": 7283513.0,
+      "step": 2750
+    },
+    {
+      "entropy": 0.4002616259455681,
+      "epoch": 6.863803680981595,
+      "grad_norm": 1.069462776184082,
+      "learning_rate": 0.0001014183434029644,
+      "loss": 0.30735193252563475,
+      "mean_token_accuracy": 0.8880347841978073,
+      "num_tokens": 7411578.0,
+      "step": 2800
+    },
+    {
+      "entropy": 0.3951660317182541,
+      "epoch": 6.986503067484662,
+      "grad_norm": 0.8845367431640625,
+      "learning_rate": 9.439032818204105e-05,
+      "loss": 0.30523622512817383,
+      "mean_token_accuracy": 0.8894886714220047,
+      "num_tokens": 7548356.0,
+      "step": 2850
+    },
+    {
+      "epoch": 7.0,
+      "eval_entropy": 0.4834100874832698,
+      "eval_mean_token_accuracy": 0.788761088166918,
+      "eval_not_syn_loss": 0.7379753589630127,
+      "eval_not_syn_runtime": 55.6309,
+      "eval_not_syn_samples_per_second": 25.076,
+      "eval_not_syn_steps_per_second": 3.146,
+      "eval_num_tokens": 7561372.0,
+      "step": 2856
+    },
+    {
+      "epoch": 7.0,
+      "eval_entropy": 0.453818735906056,
+      "eval_mean_token_accuracy": 0.8209771479879107,
+      "eval_num_tokens": 7561372.0,
+      "eval_syn_loss": 0.6787521243095398,
+      "eval_syn_runtime": 56.9966,
+      "eval_syn_samples_per_second": 24.475,
+      "eval_syn_steps_per_second": 3.07,
+      "step": 2856
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 4080,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.5044315164249766e+17,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3264/README.md b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3264/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..7ac50e2706e5b4eae8f28e204601afb241c49e55
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3264/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-2B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-2B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3264/adapter_config.json b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3264/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b9596096c71017d49955136a4e23d9f97138d305
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3264/adapter_config.json
@@ -0,0 +1,48 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.08602048083239339,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "q_proj",
+    "k_proj",
+    "gate_proj",
+    "up_proj",
+    "o_proj",
+    "down_proj",
+    "v_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3264/chat_template.jinja b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3264/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..0ef09f214eaa6d9bca297988afc1454b5827b2c7
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3264/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is true %}
+        {{- '<think>\n' }}
+    {%- else %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3264/tokenizer_config.json b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3264/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3264/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3264/trainer_state.json b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3264/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..c6a4032dd04b9b5d8eec85293573df57860c9081
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3264/trainer_state.json
@@ -0,0 +1,860 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.0,
+  "eval_steps": 500,
+  "global_step": 3264,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.2408065378665922,
+      "epoch": 0.12269938650306748,
+      "grad_norm": 1.6718658208847046,
+      "learning_rate": 4.487804970912036e-05,
+      "loss": 2.129624786376953,
+      "mean_token_accuracy": 0.5743644836544991,
+      "num_tokens": 134329.0,
+      "step": 50
+    },
+    {
+      "entropy": 1.1078504550457,
+      "epoch": 0.24539877300613497,
+      "grad_norm": 2.4091813564300537,
+      "learning_rate": 9.067197798373296e-05,
+      "loss": 1.0346390533447265,
+      "mean_token_accuracy": 0.7085381114482879,
+      "num_tokens": 262407.0,
+      "step": 100
+    },
+    {
+      "entropy": 0.9421598988771439,
+      "epoch": 0.36809815950920244,
+      "grad_norm": 1.3913646936416626,
+      "learning_rate": 0.00013646590625834558,
+      "loss": 0.8716551208496094,
+      "mean_token_accuracy": 0.7449040985107422,
+      "num_tokens": 401136.0,
+      "step": 150
+    },
+    {
+      "entropy": 0.8723550146818161,
+      "epoch": 0.49079754601226994,
+      "grad_norm": 1.1664468050003052,
+      "learning_rate": 0.0001822598345329582,
+      "loss": 0.8095977783203125,
+      "mean_token_accuracy": 0.7585240352153778,
+      "num_tokens": 531647.0,
+      "step": 200
+    },
+    {
+      "entropy": 0.8466519421339035,
+      "epoch": 0.6134969325153374,
+      "grad_norm": 1.0145421028137207,
+      "learning_rate": 0.00022805376280757083,
+      "loss": 0.7830857849121093,
+      "mean_token_accuracy": 0.7661705195903779,
+      "num_tokens": 666127.0,
+      "step": 250
+    },
+    {
+      "entropy": 0.8180295366048813,
+      "epoch": 0.7361963190184049,
+      "grad_norm": 0.9731705784797668,
+      "learning_rate": 0.00027384769108218336,
+      "loss": 0.7573603820800782,
+      "mean_token_accuracy": 0.7704453605413437,
+      "num_tokens": 799292.0,
+      "step": 300
+    },
+    {
+      "entropy": 0.7990227991342544,
+      "epoch": 0.8588957055214724,
+      "grad_norm": 1.0642374753952026,
+      "learning_rate": 0.000319641619356796,
+      "loss": 0.73709228515625,
+      "mean_token_accuracy": 0.7752733880281448,
+      "num_tokens": 930614.0,
+      "step": 350
+    },
+    {
+      "entropy": 0.7899087685346603,
+      "epoch": 0.9815950920245399,
+      "grad_norm": 0.880613386631012,
+      "learning_rate": 0.0003654355476314086,
+      "loss": 0.7347319793701171,
+      "mean_token_accuracy": 0.7766731631755829,
+      "num_tokens": 1061654.0,
+      "step": 400
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.83522472347532,
+      "eval_mean_token_accuracy": 0.7484098076820374,
+      "eval_not_syn_loss": 0.7872886061668396,
+      "eval_not_syn_runtime": 54.8697,
+      "eval_not_syn_samples_per_second": 25.424,
+      "eval_not_syn_steps_per_second": 3.189,
+      "eval_num_tokens": 1080196.0,
+      "step": 408
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.7784442271505083,
+      "eval_mean_token_accuracy": 0.8021791185651507,
+      "eval_num_tokens": 1080196.0,
+      "eval_syn_loss": 0.6980098485946655,
+      "eval_syn_runtime": 56.6342,
+      "eval_syn_samples_per_second": 24.632,
+      "eval_syn_steps_per_second": 3.09,
+      "step": 408
+    },
+    {
+      "entropy": 0.75672573694075,
+      "epoch": 1.1030674846625768,
+      "grad_norm": 0.9244198203086853,
+      "learning_rate": 0.00037356351883456205,
+      "loss": 0.7048745727539063,
+      "mean_token_accuracy": 0.783016896609104,
+      "num_tokens": 1199170.0,
+      "step": 450
+    },
+    {
+      "entropy": 0.7406797379255294,
+      "epoch": 1.2257668711656442,
+      "grad_norm": 0.6744896769523621,
+      "learning_rate": 0.00037311248151765587,
+      "loss": 0.6860730743408203,
+      "mean_token_accuracy": 0.787369327545166,
+      "num_tokens": 1334298.0,
+      "step": 500
+    },
+    {
+      "entropy": 0.7424973237514496,
+      "epoch": 1.3484662576687116,
+      "grad_norm": 0.92305588722229,
+      "learning_rate": 0.000372320629215228,
+      "loss": 0.6855724334716797,
+      "mean_token_accuracy": 0.7871134179830551,
+      "num_tokens": 1463189.0,
+      "step": 550
+    },
+    {
+      "entropy": 0.7502243828773498,
+      "epoch": 1.471165644171779,
+      "grad_norm": 0.8259047269821167,
+      "learning_rate": 0.00037118941074037944,
+      "loss": 0.6949834442138672,
+      "mean_token_accuracy": 0.7851914083957672,
+      "num_tokens": 1590310.0,
+      "step": 600
+    },
+    {
+      "entropy": 0.7233260518312454,
+      "epoch": 1.5938650306748468,
+      "grad_norm": 0.882722795009613,
+      "learning_rate": 0.00036972089582775814,
+      "loss": 0.6649005889892579,
+      "mean_token_accuracy": 0.7898879665136337,
+      "num_tokens": 1721836.0,
+      "step": 650
+    },
+    {
+      "entropy": 0.7162708967924118,
+      "epoch": 1.716564417177914,
+      "grad_norm": 0.6593027114868164,
+      "learning_rate": 0.0003679177713466678,
+      "loss": 0.6653710174560546,
+      "mean_token_accuracy": 0.7905678844451904,
+      "num_tokens": 1853507.0,
+      "step": 700
+    },
+    {
+      "entropy": 0.7150520020723343,
+      "epoch": 1.8392638036809816,
+      "grad_norm": 0.665340781211853,
+      "learning_rate": 0.0003657833363850354,
+      "loss": 0.6600938415527344,
+      "mean_token_accuracy": 0.7918649202585221,
+      "num_tokens": 1988113.0,
+      "step": 750
+    },
+    {
+      "entropy": 0.6993386310338974,
+      "epoch": 1.961963190184049,
+      "grad_norm": 0.6502187252044678,
+      "learning_rate": 0.00036332149621323294,
+      "loss": 0.6469607543945313,
+      "mean_token_accuracy": 0.7950778317451477,
+      "num_tokens": 2121314.0,
+      "step": 800
+    },
+    {
+      "epoch": 2.0,
+      "eval_entropy": 0.7315139893123082,
+      "eval_mean_token_accuracy": 0.7713504004478454,
+      "eval_not_syn_loss": 0.699893593788147,
+      "eval_not_syn_runtime": 55.4135,
+      "eval_not_syn_samples_per_second": 25.174,
+      "eval_not_syn_steps_per_second": 3.158,
+      "eval_num_tokens": 2160392.0,
+      "step": 816
+    },
+    {
+      "epoch": 2.0,
+      "eval_entropy": 0.6848451798302787,
+      "eval_mean_token_accuracy": 0.814895657471248,
+      "eval_num_tokens": 2160392.0,
+      "eval_syn_loss": 0.6415576934814453,
+      "eval_syn_runtime": 56.4735,
+      "eval_syn_samples_per_second": 24.702,
+      "eval_syn_steps_per_second": 3.099,
+      "step": 816
+    },
+    {
+      "entropy": 0.6596363644407253,
+      "epoch": 2.083435582822086,
+      "grad_norm": 0.6647700071334839,
+      "learning_rate": 0.00036053675513879666,
+      "loss": 0.599151611328125,
+      "mean_token_accuracy": 0.8071986864311527,
+      "num_tokens": 2253179.0,
+      "step": 850
+    },
+    {
+      "entropy": 0.6626429110765457,
+      "epoch": 2.2061349693251535,
+      "grad_norm": 0.7356868386268616,
+      "learning_rate": 0.00035743420826511783,
+      "loss": 0.6004195404052735,
+      "mean_token_accuracy": 0.805362731218338,
+      "num_tokens": 2379965.0,
+      "step": 900
+    },
+    {
+      "entropy": 0.6503470712900161,
+      "epoch": 2.3288343558282207,
+      "grad_norm": 1.095608115196228,
+      "learning_rate": 0.0003540195321691835,
+      "loss": 0.5951333999633789,
+      "mean_token_accuracy": 0.807324025630951,
+      "num_tokens": 2515593.0,
+      "step": 950
+    },
+    {
+      "entropy": 0.660577870607376,
+      "epoch": 2.4515337423312884,
+      "grad_norm": 0.6602179408073425,
+      "learning_rate": 0.00035029897451542325,
+      "loss": 0.6056365585327148,
+      "mean_token_accuracy": 0.8057438349723816,
+      "num_tokens": 2642790.0,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6555853113532066,
+      "epoch": 2.574233128834356,
+      "grad_norm": 0.7459151148796082,
+      "learning_rate": 0.00034627934262466606,
+      "loss": 0.5963204193115235,
+      "mean_token_accuracy": 0.8060598099231719,
+      "num_tokens": 2774396.0,
+      "step": 1050
+    },
+    {
+      "entropy": 0.6505853188037872,
+      "epoch": 2.6969325153374233,
+      "grad_norm": 0.6359792947769165,
+      "learning_rate": 0.00034196799101912103,
+      "loss": 0.5993065643310547,
+      "mean_token_accuracy": 0.806280387043953,
+      "num_tokens": 2912102.0,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6509970253705979,
+      "epoch": 2.819631901840491,
+      "grad_norm": 0.8086174726486206,
+      "learning_rate": 0.0003373728079661707,
+      "loss": 0.6001268005371094,
+      "mean_token_accuracy": 0.806866540312767,
+      "num_tokens": 3042987.0,
+      "step": 1150
+    },
+    {
+      "entropy": 0.6404844307899475,
+      "epoch": 2.942331288343558,
+      "grad_norm": 0.7038407325744629,
+      "learning_rate": 0.0003325022010455975,
+      "loss": 0.5870526123046875,
+      "mean_token_accuracy": 0.8104586935043335,
+      "num_tokens": 3178044.0,
+      "step": 1200
+    },
+    {
+      "epoch": 3.0,
+      "eval_entropy": 0.6645996727262224,
+      "eval_mean_token_accuracy": 0.798654237134116,
+      "eval_not_syn_loss": 0.6586771011352539,
+      "eval_not_syn_runtime": 55.1088,
+      "eval_not_syn_samples_per_second": 25.314,
+      "eval_not_syn_steps_per_second": 3.176,
+      "eval_num_tokens": 3240588.0,
+      "step": 1224
+    },
+    {
+      "epoch": 3.0,
+      "eval_entropy": 0.6233607809884207,
+      "eval_mean_token_accuracy": 0.7977249867575509,
+      "eval_num_tokens": 3240588.0,
+      "eval_syn_loss": 0.6288298964500427,
+      "eval_syn_runtime": 56.6081,
+      "eval_syn_samples_per_second": 24.643,
+      "eval_syn_steps_per_second": 3.091,
+      "step": 1224
+    },
+    {
+      "entropy": 0.6081961104364106,
+      "epoch": 3.063803680981595,
+      "grad_norm": 0.6161203980445862,
+      "learning_rate": 0.0003273650817666497,
+      "loss": 0.5481716537475586,
+      "mean_token_accuracy": 0.8175542035488167,
+      "num_tokens": 3312797.0,
+      "step": 1250
+    },
+    {
+      "entropy": 0.5877064895629883,
+      "epoch": 3.1865030674846624,
+      "grad_norm": 0.6909636855125427,
+      "learning_rate": 0.00032197084926309234,
+      "loss": 0.5224150466918945,
+      "mean_token_accuracy": 0.8250984263420105,
+      "num_tokens": 3441629.0,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5952656552195549,
+      "epoch": 3.30920245398773,
+      "grad_norm": 0.8147026300430298,
+      "learning_rate": 0.0003163293730960758,
+      "loss": 0.5349474334716797,
+      "mean_token_accuracy": 0.8213064837455749,
+      "num_tokens": 3576093.0,
+      "step": 1350
+    },
+    {
+      "entropy": 0.5937123569846153,
+      "epoch": 3.4319018404907977,
+      "grad_norm": 0.8690970540046692,
+      "learning_rate": 0.0003104509751962869,
+      "loss": 0.5310481262207031,
+      "mean_token_accuracy": 0.8229135835170746,
+      "num_tokens": 3705912.0,
+      "step": 1400
+    },
+    {
+      "entropy": 0.6067730465531349,
+      "epoch": 3.554601226993865,
+      "grad_norm": 0.7433330416679382,
+      "learning_rate": 0.0003043464109784207,
+      "loss": 0.5450045394897461,
+      "mean_token_accuracy": 0.8205251175165177,
+      "num_tokens": 3832376.0,
+      "step": 1450
+    },
+    {
+      "entropy": 0.5926914718747139,
+      "epoch": 3.6773006134969326,
+      "grad_norm": 0.7265587449073792,
+      "learning_rate": 0.00029802684966252974,
+      "loss": 0.5330168914794922,
+      "mean_token_accuracy": 0.821294127702713,
+      "num_tokens": 3967273.0,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5881199997663498,
+      "epoch": 3.8,
+      "grad_norm": 0.7681817412376404,
+      "learning_rate": 0.00029150385383825187,
+      "loss": 0.5288929367065429,
+      "mean_token_accuracy": 0.8234161508083343,
+      "num_tokens": 4102434.0,
+      "step": 1550
+    },
+    {
+      "entropy": 0.6008686199784279,
+      "epoch": 3.9226993865030675,
+      "grad_norm": 0.9384056329727173,
+      "learning_rate": 0.0002847893583093113,
+      "loss": 0.5437511444091797,
+      "mean_token_accuracy": 0.8201366275548935,
+      "num_tokens": 4234037.0,
+      "step": 1600
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.6218148783275059,
+      "eval_mean_token_accuracy": 0.783796865599496,
+      "eval_not_syn_loss": 0.6558170318603516,
+      "eval_not_syn_runtime": 55.429,
+      "eval_not_syn_samples_per_second": 25.167,
+      "eval_not_syn_steps_per_second": 3.157,
+      "eval_num_tokens": 4320784.0,
+      "step": 1632
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5833754854542869,
+      "eval_mean_token_accuracy": 0.8278223422595433,
+      "eval_num_tokens": 4320784.0,
+      "eval_syn_loss": 0.604770302772522,
+      "eval_syn_runtime": 56.9593,
+      "eval_syn_samples_per_second": 24.491,
+      "eval_syn_steps_per_second": 3.072,
+      "step": 1632
+    },
+    {
+      "entropy": 0.559690699733869,
+      "epoch": 4.044171779141104,
+      "grad_norm": 0.65739905834198,
+      "learning_rate": 0.0002778956482569963,
+      "loss": 0.49702674865722657,
+      "mean_token_accuracy": 0.8317034166268628,
+      "num_tokens": 4368268.0,
+      "step": 1650
+    },
+    {
+      "entropy": 0.5401161375641823,
+      "epoch": 4.166871165644172,
+      "grad_norm": 1.5176353454589844,
+      "learning_rate": 0.0002708353367625691,
+      "loss": 0.4598746109008789,
+      "mean_token_accuracy": 0.8412033343315124,
+      "num_tokens": 4494525.0,
+      "step": 1700
+    },
+    {
+      "entropy": 0.5299916860461235,
+      "epoch": 4.289570552147239,
+      "grad_norm": 0.685965359210968,
+      "learning_rate": 0.0002636213417297333,
+      "loss": 0.45663665771484374,
+      "mean_token_accuracy": 0.8412413811683654,
+      "num_tokens": 4631985.0,
+      "step": 1750
+    },
+    {
+      "entropy": 0.5352088099718094,
+      "epoch": 4.412269938650307,
+      "grad_norm": 0.8471158742904663,
+      "learning_rate": 0.0002562668622493824,
+      "loss": 0.4625707626342773,
+      "mean_token_accuracy": 0.8400991547107697,
+      "num_tokens": 4766323.0,
+      "step": 1800
+    },
+    {
+      "entropy": 0.5425067806243896,
+      "epoch": 4.534969325153375,
+      "grad_norm": 0.8595437407493591,
+      "learning_rate": 0.00024878535444987474,
+      "loss": 0.47124954223632814,
+      "mean_token_accuracy": 0.8376108628511428,
+      "num_tokens": 4895606.0,
+      "step": 1850
+    },
+    {
+      "entropy": 0.5419312778115273,
+      "epoch": 4.6576687116564415,
+      "grad_norm": 0.8436959981918335,
+      "learning_rate": 0.00024119050687701818,
+      "loss": 0.4708108901977539,
+      "mean_token_accuracy": 0.8390797019004822,
+      "num_tokens": 5025912.0,
+      "step": 1900
+    },
+    {
+      "entropy": 0.5392271488904953,
+      "epoch": 4.780368098159509,
+      "grad_norm": 0.8691944479942322,
+      "learning_rate": 0.00023349621544881404,
+      "loss": 0.46682029724121094,
+      "mean_token_accuracy": 0.8385172665119172,
+      "num_tokens": 5162972.0,
+      "step": 1950
+    },
+    {
+      "entropy": 0.5332251599431038,
+      "epoch": 4.903067484662577,
+      "grad_norm": 0.6814598441123962,
+      "learning_rate": 0.00022571655803078078,
+      "loss": 0.4687493896484375,
+      "mean_token_accuracy": 0.8388861840963364,
+      "num_tokens": 5297935.0,
+      "step": 2000
+    },
+    {
+      "epoch": 5.0,
+      "eval_entropy": 0.5847777877535139,
+      "eval_mean_token_accuracy": 0.7782382583618164,
+      "eval_not_syn_loss": 0.6678160429000854,
+      "eval_not_syn_runtime": 55.1467,
+      "eval_not_syn_samples_per_second": 25.296,
+      "eval_not_syn_steps_per_second": 3.173,
+      "eval_num_tokens": 5400980.0,
+      "step": 2040
+    },
+    {
+      "epoch": 5.0,
+      "eval_entropy": 0.5505962218557086,
+      "eval_mean_token_accuracy": 0.8308270008223397,
+      "eval_num_tokens": 5400980.0,
+      "eval_syn_loss": 0.606867253780365,
+      "eval_syn_runtime": 56.9768,
+      "eval_syn_samples_per_second": 24.484,
+      "eval_syn_steps_per_second": 3.071,
+      "step": 2040
+    },
+    {
+      "entropy": 0.533086326688227,
+      "epoch": 5.024539877300613,
+      "grad_norm": 0.9453939199447632,
+      "learning_rate": 0.00021786576867837776,
+      "loss": 0.45802555084228513,
+      "mean_token_accuracy": 0.8429379854539428,
+      "num_tokens": 5427362.0,
+      "step": 2050
+    },
+    {
+      "entropy": 0.46721911430358887,
+      "epoch": 5.147239263803681,
+      "grad_norm": 0.9237910509109497,
+      "learning_rate": 0.00020995821159365647,
+      "loss": 0.37914146423339845,
+      "mean_token_accuracy": 0.8648965793848038,
+      "num_tokens": 5564186.0,
+      "step": 2100
+    },
+    {
+      "entropy": 0.4682381117343903,
+      "epoch": 5.269938650306749,
+      "grad_norm": 0.7424130439758301,
+      "learning_rate": 0.0002020083548437879,
+      "loss": 0.3859384536743164,
+      "mean_token_accuracy": 0.8622810471057892,
+      "num_tokens": 5694728.0,
+      "step": 2150
+    },
+    {
+      "entropy": 0.464998793900013,
+      "epoch": 5.392638036809816,
+      "grad_norm": 0.9224902987480164,
+      "learning_rate": 0.0001940307438895544,
+      "loss": 0.3839593505859375,
+      "mean_token_accuracy": 0.8627353543043137,
+      "num_tokens": 5832581.0,
+      "step": 2200
+    },
+    {
+      "entropy": 0.4753906837105751,
+      "epoch": 5.515337423312883,
+      "grad_norm": 0.8931427001953125,
+      "learning_rate": 0.0001860399749722367,
+      "loss": 0.393004264831543,
+      "mean_token_accuracy": 0.8586806547641754,
+      "num_tokens": 5962417.0,
+      "step": 2250
+    },
+    {
+      "entropy": 0.4671635267138481,
+      "epoch": 5.638036809815951,
+      "grad_norm": 0.9559172987937927,
+      "learning_rate": 0.00017805066840759193,
+      "loss": 0.38852767944335936,
+      "mean_token_accuracy": 0.8615026396512985,
+      "num_tokens": 6092338.0,
+      "step": 2300
+    },
+    {
+      "entropy": 0.4746100226044655,
+      "epoch": 5.7607361963190185,
+      "grad_norm": 0.9070253968238831,
+      "learning_rate": 0.00017007744183578254,
+      "loss": 0.3960166168212891,
+      "mean_token_accuracy": 0.8591635781526565,
+      "num_tokens": 6226327.0,
+      "step": 2350
+    },
+    {
+      "entropy": 0.4698092398047447,
+      "epoch": 5.883435582822086,
+      "grad_norm": 0.8354232907295227,
+      "learning_rate": 0.0001621348834762011,
+      "loss": 0.39321571350097656,
+      "mean_token_accuracy": 0.8603039687871933,
+      "num_tokens": 6359083.0,
+      "step": 2400
+    },
+    {
+      "epoch": 6.0,
+      "eval_entropy": 0.536405998979296,
+      "eval_mean_token_accuracy": 0.790239520413535,
+      "eval_not_syn_loss": 0.6884409785270691,
+      "eval_not_syn_runtime": 54.9927,
+      "eval_not_syn_samples_per_second": 25.367,
+      "eval_not_syn_steps_per_second": 3.182,
+      "eval_num_tokens": 6481176.0,
+      "step": 2448
+    },
+    {
+      "epoch": 6.0,
+      "eval_entropy": 0.5039791010107313,
+      "eval_mean_token_accuracy": 0.8163431504794529,
+      "eval_num_tokens": 6481176.0,
+      "eval_syn_loss": 0.6377155184745789,
+      "eval_syn_runtime": 57.196,
+      "eval_syn_samples_per_second": 24.39,
+      "eval_syn_steps_per_second": 3.06,
+      "step": 2448
+    },
+    {
+      "entropy": 0.4671963176341972,
+      "epoch": 6.004907975460123,
+      "grad_norm": 0.8445424437522888,
+      "learning_rate": 0.00015423752543612516,
+      "loss": 0.3872825241088867,
+      "mean_token_accuracy": 0.8618519083418027,
+      "num_tokens": 6486549.0,
+      "step": 2450
+    },
+    {
+      "entropy": 0.3890652832388878,
+      "epoch": 6.12760736196319,
+      "grad_norm": 0.725812554359436,
+      "learning_rate": 0.00014639981712203683,
+      "loss": 0.2923568153381348,
+      "mean_token_accuracy": 0.8928290349245072,
+      "num_tokens": 6622384.0,
+      "step": 2500
+    },
+    {
+      "entropy": 0.39649384289979933,
+      "epoch": 6.250306748466258,
+      "grad_norm": 0.9143152236938477,
+      "learning_rate": 0.000138636098802257,
+      "loss": 0.303057861328125,
+      "mean_token_accuracy": 0.8888710218667984,
+      "num_tokens": 6758637.0,
+      "step": 2550
+    },
+    {
+      "entropy": 0.4020465177297592,
+      "epoch": 6.373006134969325,
+      "grad_norm": 0.9407342076301575,
+      "learning_rate": 0.00013096057536926304,
+      "loss": 0.3073894691467285,
+      "mean_token_accuracy": 0.8894575881958008,
+      "num_tokens": 6888533.0,
+      "step": 2600
+    },
+    {
+      "entropy": 0.397317076921463,
+      "epoch": 6.495705521472392,
+      "grad_norm": 0.8517777323722839,
+      "learning_rate": 0.00012338729034969828,
+      "loss": 0.30508209228515626,
+      "mean_token_accuracy": 0.8886856293678284,
+      "num_tokens": 7023729.0,
+      "step": 2650
+    },
+    {
+      "entropy": 0.3915065750479698,
+      "epoch": 6.61840490797546,
+      "grad_norm": 0.862319827079773,
+      "learning_rate": 0.00011593010020962378,
+      "loss": 0.30237462997436526,
+      "mean_token_accuracy": 0.8897094148397445,
+      "num_tokens": 7153821.0,
+      "step": 2700
+    },
+    {
+      "entropy": 0.3977497002482414,
+      "epoch": 6.741104294478528,
+      "grad_norm": 0.8821370601654053,
+      "learning_rate": 0.00010860264900202563,
+      "loss": 0.3067088508605957,
+      "mean_token_accuracy": 0.8873285222053527,
+      "num_tokens": 7283513.0,
+      "step": 2750
+    },
+    {
+      "entropy": 0.4002616259455681,
+      "epoch": 6.863803680981595,
+      "grad_norm": 1.069462776184082,
+      "learning_rate": 0.0001014183434029644,
+      "loss": 0.30735193252563475,
+      "mean_token_accuracy": 0.8880347841978073,
+      "num_tokens": 7411578.0,
+      "step": 2800
+    },
+    {
+      "entropy": 0.3951660317182541,
+      "epoch": 6.986503067484662,
+      "grad_norm": 0.8845367431640625,
+      "learning_rate": 9.439032818204105e-05,
+      "loss": 0.30523622512817383,
+      "mean_token_accuracy": 0.8894886714220047,
+      "num_tokens": 7548356.0,
+      "step": 2850
+    },
+    {
+      "epoch": 7.0,
+      "eval_entropy": 0.4834100874832698,
+      "eval_mean_token_accuracy": 0.788761088166918,
+      "eval_not_syn_loss": 0.7379753589630127,
+      "eval_not_syn_runtime": 55.6309,
+      "eval_not_syn_samples_per_second": 25.076,
+      "eval_not_syn_steps_per_second": 3.146,
+      "eval_num_tokens": 7561372.0,
+      "step": 2856
+    },
+    {
+      "epoch": 7.0,
+      "eval_entropy": 0.453818735906056,
+      "eval_mean_token_accuracy": 0.8209771479879107,
+      "eval_num_tokens": 7561372.0,
+      "eval_syn_loss": 0.6787521243095398,
+      "eval_syn_runtime": 56.9966,
+      "eval_syn_samples_per_second": 24.475,
+      "eval_syn_steps_per_second": 3.07,
+      "step": 2856
+    },
+    {
+      "entropy": 0.3386827930055483,
+      "epoch": 7.1079754601227,
+      "grad_norm": 1.0214653015136719,
+      "learning_rate": 8.753146215206042e-05,
+      "loss": 0.2361087417602539,
+      "mean_token_accuracy": 0.9141153067049353,
+      "num_tokens": 7671657.0,
+      "step": 2900
+    },
+    {
+      "entropy": 0.31863692075014116,
+      "epoch": 7.230674846625767,
+      "grad_norm": 0.9141477346420288,
+      "learning_rate": 8.085429464189586e-05,
+      "loss": 0.21914518356323243,
+      "mean_token_accuracy": 0.9194145703315735,
+      "num_tokens": 7806989.0,
+      "step": 2950
+    },
+    {
+      "entropy": 0.31718266174197196,
+      "epoch": 7.353374233128834,
+      "grad_norm": 0.9127821922302246,
+      "learning_rate": 7.437104253560078e-05,
+      "loss": 0.22112932205200195,
+      "mean_token_accuracy": 0.9190003037452698,
+      "num_tokens": 7941760.0,
+      "step": 3000
+    },
+    {
+      "entropy": 0.3143314927816391,
+      "epoch": 7.476073619631902,
+      "grad_norm": 1.1398990154266357,
+      "learning_rate": 6.809356791977874e-05,
+      "loss": 0.2191557502746582,
+      "mean_token_accuracy": 0.918931410908699,
+      "num_tokens": 8080328.0,
+      "step": 3050
+    },
+    {
+      "entropy": 0.32102111548185347,
+      "epoch": 7.598773006134969,
+      "grad_norm": 1.069332480430603,
+      "learning_rate": 6.203335638010852e-05,
+      "loss": 0.22624496459960938,
+      "mean_token_accuracy": 0.9173947256803513,
+      "num_tokens": 8210831.0,
+      "step": 3100
+    },
+    {
+      "entropy": 0.31791565626859664,
+      "epoch": 7.721472392638037,
+      "grad_norm": 1.1536624431610107,
+      "learning_rate": 5.62014959867348e-05,
+      "loss": 0.22338006973266603,
+      "mean_token_accuracy": 0.9177713346481323,
+      "num_tokens": 8344985.0,
+      "step": 3150
+    },
+    {
+      "entropy": 0.3209437537193298,
+      "epoch": 7.844171779141105,
+      "grad_norm": 1.1040980815887451,
+      "learning_rate": 5.06086570069728e-05,
+      "loss": 0.2263322639465332,
+      "mean_token_accuracy": 0.9170012313127518,
+      "num_tokens": 8475716.0,
+      "step": 3200
+    },
+    {
+      "entropy": 0.31527657777071,
+      "epoch": 7.9668711656441715,
+      "grad_norm": 1.1415972709655762,
+      "learning_rate": 4.526507238244653e-05,
+      "loss": 0.2216329574584961,
+      "mean_token_accuracy": 0.918340762257576,
+      "num_tokens": 8606709.0,
+      "step": 3250
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.43602277857916694,
+      "eval_mean_token_accuracy": 0.7907402624402727,
+      "eval_not_syn_loss": 0.8163036704063416,
+      "eval_not_syn_runtime": 55.3868,
+      "eval_not_syn_samples_per_second": 25.187,
+      "eval_not_syn_steps_per_second": 3.16,
+      "eval_num_tokens": 8641568.0,
+      "step": 3264
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.4093018649305616,
+      "eval_mean_token_accuracy": 0.8147984709058489,
+      "eval_num_tokens": 8641568.0,
+      "eval_syn_loss": 0.7571278810501099,
+      "eval_syn_runtime": 56.797,
+      "eval_syn_samples_per_second": 24.561,
+      "eval_syn_steps_per_second": 3.081,
+      "step": 3264
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 4080,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.7177733506397658e+17,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3672/README.md b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3672/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..7ac50e2706e5b4eae8f28e204601afb241c49e55
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3672/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-2B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-2B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3672/adapter_config.json b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3672/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b9596096c71017d49955136a4e23d9f97138d305
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3672/adapter_config.json
@@ -0,0 +1,48 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.08602048083239339,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "q_proj",
+    "k_proj",
+    "gate_proj",
+    "up_proj",
+    "o_proj",
+    "down_proj",
+    "v_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3672/chat_template.jinja b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3672/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..0ef09f214eaa6d9bca297988afc1454b5827b2c7
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3672/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is true %}
+        {{- '<think>\n' }}
+    {%- else %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3672/tokenizer_config.json b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3672/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3672/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3672/trainer_state.json b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3672/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..193e38e9f9417f76bd973295aa24eba60d0fdef6
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-3672/trainer_state.json
@@ -0,0 +1,962 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.0,
+  "eval_steps": 500,
+  "global_step": 3672,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.2408065378665922,
+      "epoch": 0.12269938650306748,
+      "grad_norm": 1.6718658208847046,
+      "learning_rate": 4.487804970912036e-05,
+      "loss": 2.129624786376953,
+      "mean_token_accuracy": 0.5743644836544991,
+      "num_tokens": 134329.0,
+      "step": 50
+    },
+    {
+      "entropy": 1.1078504550457,
+      "epoch": 0.24539877300613497,
+      "grad_norm": 2.4091813564300537,
+      "learning_rate": 9.067197798373296e-05,
+      "loss": 1.0346390533447265,
+      "mean_token_accuracy": 0.7085381114482879,
+      "num_tokens": 262407.0,
+      "step": 100
+    },
+    {
+      "entropy": 0.9421598988771439,
+      "epoch": 0.36809815950920244,
+      "grad_norm": 1.3913646936416626,
+      "learning_rate": 0.00013646590625834558,
+      "loss": 0.8716551208496094,
+      "mean_token_accuracy": 0.7449040985107422,
+      "num_tokens": 401136.0,
+      "step": 150
+    },
+    {
+      "entropy": 0.8723550146818161,
+      "epoch": 0.49079754601226994,
+      "grad_norm": 1.1664468050003052,
+      "learning_rate": 0.0001822598345329582,
+      "loss": 0.8095977783203125,
+      "mean_token_accuracy": 0.7585240352153778,
+      "num_tokens": 531647.0,
+      "step": 200
+    },
+    {
+      "entropy": 0.8466519421339035,
+      "epoch": 0.6134969325153374,
+      "grad_norm": 1.0145421028137207,
+      "learning_rate": 0.00022805376280757083,
+      "loss": 0.7830857849121093,
+      "mean_token_accuracy": 0.7661705195903779,
+      "num_tokens": 666127.0,
+      "step": 250
+    },
+    {
+      "entropy": 0.8180295366048813,
+      "epoch": 0.7361963190184049,
+      "grad_norm": 0.9731705784797668,
+      "learning_rate": 0.00027384769108218336,
+      "loss": 0.7573603820800782,
+      "mean_token_accuracy": 0.7704453605413437,
+      "num_tokens": 799292.0,
+      "step": 300
+    },
+    {
+      "entropy": 0.7990227991342544,
+      "epoch": 0.8588957055214724,
+      "grad_norm": 1.0642374753952026,
+      "learning_rate": 0.000319641619356796,
+      "loss": 0.73709228515625,
+      "mean_token_accuracy": 0.7752733880281448,
+      "num_tokens": 930614.0,
+      "step": 350
+    },
+    {
+      "entropy": 0.7899087685346603,
+      "epoch": 0.9815950920245399,
+      "grad_norm": 0.880613386631012,
+      "learning_rate": 0.0003654355476314086,
+      "loss": 0.7347319793701171,
+      "mean_token_accuracy": 0.7766731631755829,
+      "num_tokens": 1061654.0,
+      "step": 400
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.83522472347532,
+      "eval_mean_token_accuracy": 0.7484098076820374,
+      "eval_not_syn_loss": 0.7872886061668396,
+      "eval_not_syn_runtime": 54.8697,
+      "eval_not_syn_samples_per_second": 25.424,
+      "eval_not_syn_steps_per_second": 3.189,
+      "eval_num_tokens": 1080196.0,
+      "step": 408
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.7784442271505083,
+      "eval_mean_token_accuracy": 0.8021791185651507,
+      "eval_num_tokens": 1080196.0,
+      "eval_syn_loss": 0.6980098485946655,
+      "eval_syn_runtime": 56.6342,
+      "eval_syn_samples_per_second": 24.632,
+      "eval_syn_steps_per_second": 3.09,
+      "step": 408
+    },
+    {
+      "entropy": 0.75672573694075,
+      "epoch": 1.1030674846625768,
+      "grad_norm": 0.9244198203086853,
+      "learning_rate": 0.00037356351883456205,
+      "loss": 0.7048745727539063,
+      "mean_token_accuracy": 0.783016896609104,
+      "num_tokens": 1199170.0,
+      "step": 450
+    },
+    {
+      "entropy": 0.7406797379255294,
+      "epoch": 1.2257668711656442,
+      "grad_norm": 0.6744896769523621,
+      "learning_rate": 0.00037311248151765587,
+      "loss": 0.6860730743408203,
+      "mean_token_accuracy": 0.787369327545166,
+      "num_tokens": 1334298.0,
+      "step": 500
+    },
+    {
+      "entropy": 0.7424973237514496,
+      "epoch": 1.3484662576687116,
+      "grad_norm": 0.92305588722229,
+      "learning_rate": 0.000372320629215228,
+      "loss": 0.6855724334716797,
+      "mean_token_accuracy": 0.7871134179830551,
+      "num_tokens": 1463189.0,
+      "step": 550
+    },
+    {
+      "entropy": 0.7502243828773498,
+      "epoch": 1.471165644171779,
+      "grad_norm": 0.8259047269821167,
+      "learning_rate": 0.00037118941074037944,
+      "loss": 0.6949834442138672,
+      "mean_token_accuracy": 0.7851914083957672,
+      "num_tokens": 1590310.0,
+      "step": 600
+    },
+    {
+      "entropy": 0.7233260518312454,
+      "epoch": 1.5938650306748468,
+      "grad_norm": 0.882722795009613,
+      "learning_rate": 0.00036972089582775814,
+      "loss": 0.6649005889892579,
+      "mean_token_accuracy": 0.7898879665136337,
+      "num_tokens": 1721836.0,
+      "step": 650
+    },
+    {
+      "entropy": 0.7162708967924118,
+      "epoch": 1.716564417177914,
+      "grad_norm": 0.6593027114868164,
+      "learning_rate": 0.0003679177713466678,
+      "loss": 0.6653710174560546,
+      "mean_token_accuracy": 0.7905678844451904,
+      "num_tokens": 1853507.0,
+      "step": 700
+    },
+    {
+      "entropy": 0.7150520020723343,
+      "epoch": 1.8392638036809816,
+      "grad_norm": 0.665340781211853,
+      "learning_rate": 0.0003657833363850354,
+      "loss": 0.6600938415527344,
+      "mean_token_accuracy": 0.7918649202585221,
+      "num_tokens": 1988113.0,
+      "step": 750
+    },
+    {
+      "entropy": 0.6993386310338974,
+      "epoch": 1.961963190184049,
+      "grad_norm": 0.6502187252044678,
+      "learning_rate": 0.00036332149621323294,
+      "loss": 0.6469607543945313,
+      "mean_token_accuracy": 0.7950778317451477,
+      "num_tokens": 2121314.0,
+      "step": 800
+    },
+    {
+      "epoch": 2.0,
+      "eval_entropy": 0.7315139893123082,
+      "eval_mean_token_accuracy": 0.7713504004478454,
+      "eval_not_syn_loss": 0.699893593788147,
+      "eval_not_syn_runtime": 55.4135,
+      "eval_not_syn_samples_per_second": 25.174,
+      "eval_not_syn_steps_per_second": 3.158,
+      "eval_num_tokens": 2160392.0,
+      "step": 816
+    },
+    {
+      "epoch": 2.0,
+      "eval_entropy": 0.6848451798302787,
+      "eval_mean_token_accuracy": 0.814895657471248,
+      "eval_num_tokens": 2160392.0,
+      "eval_syn_loss": 0.6415576934814453,
+      "eval_syn_runtime": 56.4735,
+      "eval_syn_samples_per_second": 24.702,
+      "eval_syn_steps_per_second": 3.099,
+      "step": 816
+    },
+    {
+      "entropy": 0.6596363644407253,
+      "epoch": 2.083435582822086,
+      "grad_norm": 0.6647700071334839,
+      "learning_rate": 0.00036053675513879666,
+      "loss": 0.599151611328125,
+      "mean_token_accuracy": 0.8071986864311527,
+      "num_tokens": 2253179.0,
+      "step": 850
+    },
+    {
+      "entropy": 0.6626429110765457,
+      "epoch": 2.2061349693251535,
+      "grad_norm": 0.7356868386268616,
+      "learning_rate": 0.00035743420826511783,
+      "loss": 0.6004195404052735,
+      "mean_token_accuracy": 0.805362731218338,
+      "num_tokens": 2379965.0,
+      "step": 900
+    },
+    {
+      "entropy": 0.6503470712900161,
+      "epoch": 2.3288343558282207,
+      "grad_norm": 1.095608115196228,
+      "learning_rate": 0.0003540195321691835,
+      "loss": 0.5951333999633789,
+      "mean_token_accuracy": 0.807324025630951,
+      "num_tokens": 2515593.0,
+      "step": 950
+    },
+    {
+      "entropy": 0.660577870607376,
+      "epoch": 2.4515337423312884,
+      "grad_norm": 0.6602179408073425,
+      "learning_rate": 0.00035029897451542325,
+      "loss": 0.6056365585327148,
+      "mean_token_accuracy": 0.8057438349723816,
+      "num_tokens": 2642790.0,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6555853113532066,
+      "epoch": 2.574233128834356,
+      "grad_norm": 0.7459151148796082,
+      "learning_rate": 0.00034627934262466606,
+      "loss": 0.5963204193115235,
+      "mean_token_accuracy": 0.8060598099231719,
+      "num_tokens": 2774396.0,
+      "step": 1050
+    },
+    {
+      "entropy": 0.6505853188037872,
+      "epoch": 2.6969325153374233,
+      "grad_norm": 0.6359792947769165,
+      "learning_rate": 0.00034196799101912103,
+      "loss": 0.5993065643310547,
+      "mean_token_accuracy": 0.806280387043953,
+      "num_tokens": 2912102.0,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6509970253705979,
+      "epoch": 2.819631901840491,
+      "grad_norm": 0.8086174726486206,
+      "learning_rate": 0.0003373728079661707,
+      "loss": 0.6001268005371094,
+      "mean_token_accuracy": 0.806866540312767,
+      "num_tokens": 3042987.0,
+      "step": 1150
+    },
+    {
+      "entropy": 0.6404844307899475,
+      "epoch": 2.942331288343558,
+      "grad_norm": 0.7038407325744629,
+      "learning_rate": 0.0003325022010455975,
+      "loss": 0.5870526123046875,
+      "mean_token_accuracy": 0.8104586935043335,
+      "num_tokens": 3178044.0,
+      "step": 1200
+    },
+    {
+      "epoch": 3.0,
+      "eval_entropy": 0.6645996727262224,
+      "eval_mean_token_accuracy": 0.798654237134116,
+      "eval_not_syn_loss": 0.6586771011352539,
+      "eval_not_syn_runtime": 55.1088,
+      "eval_not_syn_samples_per_second": 25.314,
+      "eval_not_syn_steps_per_second": 3.176,
+      "eval_num_tokens": 3240588.0,
+      "step": 1224
+    },
+    {
+      "epoch": 3.0,
+      "eval_entropy": 0.6233607809884207,
+      "eval_mean_token_accuracy": 0.7977249867575509,
+      "eval_num_tokens": 3240588.0,
+      "eval_syn_loss": 0.6288298964500427,
+      "eval_syn_runtime": 56.6081,
+      "eval_syn_samples_per_second": 24.643,
+      "eval_syn_steps_per_second": 3.091,
+      "step": 1224
+    },
+    {
+      "entropy": 0.6081961104364106,
+      "epoch": 3.063803680981595,
+      "grad_norm": 0.6161203980445862,
+      "learning_rate": 0.0003273650817666497,
+      "loss": 0.5481716537475586,
+      "mean_token_accuracy": 0.8175542035488167,
+      "num_tokens": 3312797.0,
+      "step": 1250
+    },
+    {
+      "entropy": 0.5877064895629883,
+      "epoch": 3.1865030674846624,
+      "grad_norm": 0.6909636855125427,
+      "learning_rate": 0.00032197084926309234,
+      "loss": 0.5224150466918945,
+      "mean_token_accuracy": 0.8250984263420105,
+      "num_tokens": 3441629.0,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5952656552195549,
+      "epoch": 3.30920245398773,
+      "grad_norm": 0.8147026300430298,
+      "learning_rate": 0.0003163293730960758,
+      "loss": 0.5349474334716797,
+      "mean_token_accuracy": 0.8213064837455749,
+      "num_tokens": 3576093.0,
+      "step": 1350
+    },
+    {
+      "entropy": 0.5937123569846153,
+      "epoch": 3.4319018404907977,
+      "grad_norm": 0.8690970540046692,
+      "learning_rate": 0.0003104509751962869,
+      "loss": 0.5310481262207031,
+      "mean_token_accuracy": 0.8229135835170746,
+      "num_tokens": 3705912.0,
+      "step": 1400
+    },
+    {
+      "entropy": 0.6067730465531349,
+      "epoch": 3.554601226993865,
+      "grad_norm": 0.7433330416679382,
+      "learning_rate": 0.0003043464109784207,
+      "loss": 0.5450045394897461,
+      "mean_token_accuracy": 0.8205251175165177,
+      "num_tokens": 3832376.0,
+      "step": 1450
+    },
+    {
+      "entropy": 0.5926914718747139,
+      "epoch": 3.6773006134969326,
+      "grad_norm": 0.7265587449073792,
+      "learning_rate": 0.00029802684966252974,
+      "loss": 0.5330168914794922,
+      "mean_token_accuracy": 0.821294127702713,
+      "num_tokens": 3967273.0,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5881199997663498,
+      "epoch": 3.8,
+      "grad_norm": 0.7681817412376404,
+      "learning_rate": 0.00029150385383825187,
+      "loss": 0.5288929367065429,
+      "mean_token_accuracy": 0.8234161508083343,
+      "num_tokens": 4102434.0,
+      "step": 1550
+    },
+    {
+      "entropy": 0.6008686199784279,
+      "epoch": 3.9226993865030675,
+      "grad_norm": 0.9384056329727173,
+      "learning_rate": 0.0002847893583093113,
+      "loss": 0.5437511444091797,
+      "mean_token_accuracy": 0.8201366275548935,
+      "num_tokens": 4234037.0,
+      "step": 1600
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.6218148783275059,
+      "eval_mean_token_accuracy": 0.783796865599496,
+      "eval_not_syn_loss": 0.6558170318603516,
+      "eval_not_syn_runtime": 55.429,
+      "eval_not_syn_samples_per_second": 25.167,
+      "eval_not_syn_steps_per_second": 3.157,
+      "eval_num_tokens": 4320784.0,
+      "step": 1632
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5833754854542869,
+      "eval_mean_token_accuracy": 0.8278223422595433,
+      "eval_num_tokens": 4320784.0,
+      "eval_syn_loss": 0.604770302772522,
+      "eval_syn_runtime": 56.9593,
+      "eval_syn_samples_per_second": 24.491,
+      "eval_syn_steps_per_second": 3.072,
+      "step": 1632
+    },
+    {
+      "entropy": 0.559690699733869,
+      "epoch": 4.044171779141104,
+      "grad_norm": 0.65739905834198,
+      "learning_rate": 0.0002778956482569963,
+      "loss": 0.49702674865722657,
+      "mean_token_accuracy": 0.8317034166268628,
+      "num_tokens": 4368268.0,
+      "step": 1650
+    },
+    {
+      "entropy": 0.5401161375641823,
+      "epoch": 4.166871165644172,
+      "grad_norm": 1.5176353454589844,
+      "learning_rate": 0.0002708353367625691,
+      "loss": 0.4598746109008789,
+      "mean_token_accuracy": 0.8412033343315124,
+      "num_tokens": 4494525.0,
+      "step": 1700
+    },
+    {
+      "entropy": 0.5299916860461235,
+      "epoch": 4.289570552147239,
+      "grad_norm": 0.685965359210968,
+      "learning_rate": 0.0002636213417297333,
+      "loss": 0.45663665771484374,
+      "mean_token_accuracy": 0.8412413811683654,
+      "num_tokens": 4631985.0,
+      "step": 1750
+    },
+    {
+      "entropy": 0.5352088099718094,
+      "epoch": 4.412269938650307,
+      "grad_norm": 0.8471158742904663,
+      "learning_rate": 0.0002562668622493824,
+      "loss": 0.4625707626342773,
+      "mean_token_accuracy": 0.8400991547107697,
+      "num_tokens": 4766323.0,
+      "step": 1800
+    },
+    {
+      "entropy": 0.5425067806243896,
+      "epoch": 4.534969325153375,
+      "grad_norm": 0.8595437407493591,
+      "learning_rate": 0.00024878535444987474,
+      "loss": 0.47124954223632814,
+      "mean_token_accuracy": 0.8376108628511428,
+      "num_tokens": 4895606.0,
+      "step": 1850
+    },
+    {
+      "entropy": 0.5419312778115273,
+      "epoch": 4.6576687116564415,
+      "grad_norm": 0.8436959981918335,
+      "learning_rate": 0.00024119050687701818,
+      "loss": 0.4708108901977539,
+      "mean_token_accuracy": 0.8390797019004822,
+      "num_tokens": 5025912.0,
+      "step": 1900
+    },
+    {
+      "entropy": 0.5392271488904953,
+      "epoch": 4.780368098159509,
+      "grad_norm": 0.8691944479942322,
+      "learning_rate": 0.00023349621544881404,
+      "loss": 0.46682029724121094,
+      "mean_token_accuracy": 0.8385172665119172,
+      "num_tokens": 5162972.0,
+      "step": 1950
+    },
+    {
+      "entropy": 0.5332251599431038,
+      "epoch": 4.903067484662577,
+      "grad_norm": 0.6814598441123962,
+      "learning_rate": 0.00022571655803078078,
+      "loss": 0.4687493896484375,
+      "mean_token_accuracy": 0.8388861840963364,
+      "num_tokens": 5297935.0,
+      "step": 2000
+    },
+    {
+      "epoch": 5.0,
+      "eval_entropy": 0.5847777877535139,
+      "eval_mean_token_accuracy": 0.7782382583618164,
+      "eval_not_syn_loss": 0.6678160429000854,
+      "eval_not_syn_runtime": 55.1467,
+      "eval_not_syn_samples_per_second": 25.296,
+      "eval_not_syn_steps_per_second": 3.173,
+      "eval_num_tokens": 5400980.0,
+      "step": 2040
+    },
+    {
+      "epoch": 5.0,
+      "eval_entropy": 0.5505962218557086,
+      "eval_mean_token_accuracy": 0.8308270008223397,
+      "eval_num_tokens": 5400980.0,
+      "eval_syn_loss": 0.606867253780365,
+      "eval_syn_runtime": 56.9768,
+      "eval_syn_samples_per_second": 24.484,
+      "eval_syn_steps_per_second": 3.071,
+      "step": 2040
+    },
+    {
+      "entropy": 0.533086326688227,
+      "epoch": 5.024539877300613,
+      "grad_norm": 0.9453939199447632,
+      "learning_rate": 0.00021786576867837776,
+      "loss": 0.45802555084228513,
+      "mean_token_accuracy": 0.8429379854539428,
+      "num_tokens": 5427362.0,
+      "step": 2050
+    },
+    {
+      "entropy": 0.46721911430358887,
+      "epoch": 5.147239263803681,
+      "grad_norm": 0.9237910509109497,
+      "learning_rate": 0.00020995821159365647,
+      "loss": 0.37914146423339845,
+      "mean_token_accuracy": 0.8648965793848038,
+      "num_tokens": 5564186.0,
+      "step": 2100
+    },
+    {
+      "entropy": 0.4682381117343903,
+      "epoch": 5.269938650306749,
+      "grad_norm": 0.7424130439758301,
+      "learning_rate": 0.0002020083548437879,
+      "loss": 0.3859384536743164,
+      "mean_token_accuracy": 0.8622810471057892,
+      "num_tokens": 5694728.0,
+      "step": 2150
+    },
+    {
+      "entropy": 0.464998793900013,
+      "epoch": 5.392638036809816,
+      "grad_norm": 0.9224902987480164,
+      "learning_rate": 0.0001940307438895544,
+      "loss": 0.3839593505859375,
+      "mean_token_accuracy": 0.8627353543043137,
+      "num_tokens": 5832581.0,
+      "step": 2200
+    },
+    {
+      "entropy": 0.4753906837105751,
+      "epoch": 5.515337423312883,
+      "grad_norm": 0.8931427001953125,
+      "learning_rate": 0.0001860399749722367,
+      "loss": 0.393004264831543,
+      "mean_token_accuracy": 0.8586806547641754,
+      "num_tokens": 5962417.0,
+      "step": 2250
+    },
+    {
+      "entropy": 0.4671635267138481,
+      "epoch": 5.638036809815951,
+      "grad_norm": 0.9559172987937927,
+      "learning_rate": 0.00017805066840759193,
+      "loss": 0.38852767944335936,
+      "mean_token_accuracy": 0.8615026396512985,
+      "num_tokens": 6092338.0,
+      "step": 2300
+    },
+    {
+      "entropy": 0.4746100226044655,
+      "epoch": 5.7607361963190185,
+      "grad_norm": 0.9070253968238831,
+      "learning_rate": 0.00017007744183578254,
+      "loss": 0.3960166168212891,
+      "mean_token_accuracy": 0.8591635781526565,
+      "num_tokens": 6226327.0,
+      "step": 2350
+    },
+    {
+      "entropy": 0.4698092398047447,
+      "epoch": 5.883435582822086,
+      "grad_norm": 0.8354232907295227,
+      "learning_rate": 0.0001621348834762011,
+      "loss": 0.39321571350097656,
+      "mean_token_accuracy": 0.8603039687871933,
+      "num_tokens": 6359083.0,
+      "step": 2400
+    },
+    {
+      "epoch": 6.0,
+      "eval_entropy": 0.536405998979296,
+      "eval_mean_token_accuracy": 0.790239520413535,
+      "eval_not_syn_loss": 0.6884409785270691,
+      "eval_not_syn_runtime": 54.9927,
+      "eval_not_syn_samples_per_second": 25.367,
+      "eval_not_syn_steps_per_second": 3.182,
+      "eval_num_tokens": 6481176.0,
+      "step": 2448
+    },
+    {
+      "epoch": 6.0,
+      "eval_entropy": 0.5039791010107313,
+      "eval_mean_token_accuracy": 0.8163431504794529,
+      "eval_num_tokens": 6481176.0,
+      "eval_syn_loss": 0.6377155184745789,
+      "eval_syn_runtime": 57.196,
+      "eval_syn_samples_per_second": 24.39,
+      "eval_syn_steps_per_second": 3.06,
+      "step": 2448
+    },
+    {
+      "entropy": 0.4671963176341972,
+      "epoch": 6.004907975460123,
+      "grad_norm": 0.8445424437522888,
+      "learning_rate": 0.00015423752543612516,
+      "loss": 0.3872825241088867,
+      "mean_token_accuracy": 0.8618519083418027,
+      "num_tokens": 6486549.0,
+      "step": 2450
+    },
+    {
+      "entropy": 0.3890652832388878,
+      "epoch": 6.12760736196319,
+      "grad_norm": 0.725812554359436,
+      "learning_rate": 0.00014639981712203683,
+      "loss": 0.2923568153381348,
+      "mean_token_accuracy": 0.8928290349245072,
+      "num_tokens": 6622384.0,
+      "step": 2500
+    },
+    {
+      "entropy": 0.39649384289979933,
+      "epoch": 6.250306748466258,
+      "grad_norm": 0.9143152236938477,
+      "learning_rate": 0.000138636098802257,
+      "loss": 0.303057861328125,
+      "mean_token_accuracy": 0.8888710218667984,
+      "num_tokens": 6758637.0,
+      "step": 2550
+    },
+    {
+      "entropy": 0.4020465177297592,
+      "epoch": 6.373006134969325,
+      "grad_norm": 0.9407342076301575,
+      "learning_rate": 0.00013096057536926304,
+      "loss": 0.3073894691467285,
+      "mean_token_accuracy": 0.8894575881958008,
+      "num_tokens": 6888533.0,
+      "step": 2600
+    },
+    {
+      "entropy": 0.397317076921463,
+      "epoch": 6.495705521472392,
+      "grad_norm": 0.8517777323722839,
+      "learning_rate": 0.00012338729034969828,
+      "loss": 0.30508209228515626,
+      "mean_token_accuracy": 0.8886856293678284,
+      "num_tokens": 7023729.0,
+      "step": 2650
+    },
+    {
+      "entropy": 0.3915065750479698,
+      "epoch": 6.61840490797546,
+      "grad_norm": 0.862319827079773,
+      "learning_rate": 0.00011593010020962378,
+      "loss": 0.30237462997436526,
+      "mean_token_accuracy": 0.8897094148397445,
+      "num_tokens": 7153821.0,
+      "step": 2700
+    },
+    {
+      "entropy": 0.3977497002482414,
+      "epoch": 6.741104294478528,
+      "grad_norm": 0.8821370601654053,
+      "learning_rate": 0.00010860264900202563,
+      "loss": 0.3067088508605957,
+      "mean_token_accuracy": 0.8873285222053527,
+      "num_tokens": 7283513.0,
+      "step": 2750
+    },
+    {
+      "entropy": 0.4002616259455681,
+      "epoch": 6.863803680981595,
+      "grad_norm": 1.069462776184082,
+      "learning_rate": 0.0001014183434029644,
+      "loss": 0.30735193252563475,
+      "mean_token_accuracy": 0.8880347841978073,
+      "num_tokens": 7411578.0,
+      "step": 2800
+    },
+    {
+      "entropy": 0.3951660317182541,
+      "epoch": 6.986503067484662,
+      "grad_norm": 0.8845367431640625,
+      "learning_rate": 9.439032818204105e-05,
+      "loss": 0.30523622512817383,
+      "mean_token_accuracy": 0.8894886714220047,
+      "num_tokens": 7548356.0,
+      "step": 2850
+    },
+    {
+      "epoch": 7.0,
+      "eval_entropy": 0.4834100874832698,
+      "eval_mean_token_accuracy": 0.788761088166918,
+      "eval_not_syn_loss": 0.7379753589630127,
+      "eval_not_syn_runtime": 55.6309,
+      "eval_not_syn_samples_per_second": 25.076,
+      "eval_not_syn_steps_per_second": 3.146,
+      "eval_num_tokens": 7561372.0,
+      "step": 2856
+    },
+    {
+      "epoch": 7.0,
+      "eval_entropy": 0.453818735906056,
+      "eval_mean_token_accuracy": 0.8209771479879107,
+      "eval_num_tokens": 7561372.0,
+      "eval_syn_loss": 0.6787521243095398,
+      "eval_syn_runtime": 56.9966,
+      "eval_syn_samples_per_second": 24.475,
+      "eval_syn_steps_per_second": 3.07,
+      "step": 2856
+    },
+    {
+      "entropy": 0.3386827930055483,
+      "epoch": 7.1079754601227,
+      "grad_norm": 1.0214653015136719,
+      "learning_rate": 8.753146215206042e-05,
+      "loss": 0.2361087417602539,
+      "mean_token_accuracy": 0.9141153067049353,
+      "num_tokens": 7671657.0,
+      "step": 2900
+    },
+    {
+      "entropy": 0.31863692075014116,
+      "epoch": 7.230674846625767,
+      "grad_norm": 0.9141477346420288,
+      "learning_rate": 8.085429464189586e-05,
+      "loss": 0.21914518356323243,
+      "mean_token_accuracy": 0.9194145703315735,
+      "num_tokens": 7806989.0,
+      "step": 2950
+    },
+    {
+      "entropy": 0.31718266174197196,
+      "epoch": 7.353374233128834,
+      "grad_norm": 0.9127821922302246,
+      "learning_rate": 7.437104253560078e-05,
+      "loss": 0.22112932205200195,
+      "mean_token_accuracy": 0.9190003037452698,
+      "num_tokens": 7941760.0,
+      "step": 3000
+    },
+    {
+      "entropy": 0.3143314927816391,
+      "epoch": 7.476073619631902,
+      "grad_norm": 1.1398990154266357,
+      "learning_rate": 6.809356791977874e-05,
+      "loss": 0.2191557502746582,
+      "mean_token_accuracy": 0.918931410908699,
+      "num_tokens": 8080328.0,
+      "step": 3050
+    },
+    {
+      "entropy": 0.32102111548185347,
+      "epoch": 7.598773006134969,
+      "grad_norm": 1.069332480430603,
+      "learning_rate": 6.203335638010852e-05,
+      "loss": 0.22624496459960938,
+      "mean_token_accuracy": 0.9173947256803513,
+      "num_tokens": 8210831.0,
+      "step": 3100
+    },
+    {
+      "entropy": 0.31791565626859664,
+      "epoch": 7.721472392638037,
+      "grad_norm": 1.1536624431610107,
+      "learning_rate": 5.62014959867348e-05,
+      "loss": 0.22338006973266603,
+      "mean_token_accuracy": 0.9177713346481323,
+      "num_tokens": 8344985.0,
+      "step": 3150
+    },
+    {
+      "entropy": 0.3209437537193298,
+      "epoch": 7.844171779141105,
+      "grad_norm": 1.1040980815887451,
+      "learning_rate": 5.06086570069728e-05,
+      "loss": 0.2263322639465332,
+      "mean_token_accuracy": 0.9170012313127518,
+      "num_tokens": 8475716.0,
+      "step": 3200
+    },
+    {
+      "entropy": 0.31527657777071,
+      "epoch": 7.9668711656441715,
+      "grad_norm": 1.1415972709655762,
+      "learning_rate": 4.526507238244653e-05,
+      "loss": 0.2216329574584961,
+      "mean_token_accuracy": 0.918340762257576,
+      "num_tokens": 8606709.0,
+      "step": 3250
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.43602277857916694,
+      "eval_mean_token_accuracy": 0.7907402624402727,
+      "eval_not_syn_loss": 0.8163036704063416,
+      "eval_not_syn_runtime": 55.3868,
+      "eval_not_syn_samples_per_second": 25.187,
+      "eval_not_syn_steps_per_second": 3.16,
+      "eval_num_tokens": 8641568.0,
+      "step": 3264
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.4093018649305616,
+      "eval_mean_token_accuracy": 0.8147984709058489,
+      "eval_num_tokens": 8641568.0,
+      "eval_syn_loss": 0.7571278810501099,
+      "eval_syn_runtime": 56.797,
+      "eval_syn_samples_per_second": 24.561,
+      "eval_syn_steps_per_second": 3.081,
+      "step": 3264
+    },
+    {
+      "entropy": 0.2847619162063406,
+      "epoch": 8.088343558282208,
+      "grad_norm": 1.0268603563308716,
+      "learning_rate": 4.018051900638024e-05,
+      "loss": 0.1806371307373047,
+      "mean_token_accuracy": 0.9346237525795446,
+      "num_tokens": 8733256.0,
+      "step": 3300
+    },
+    {
+      "entropy": 0.26448936730623246,
+      "epoch": 8.211042944785277,
+      "grad_norm": 1.307308554649353,
+      "learning_rate": 3.536429983529935e-05,
+      "loss": 0.16438600540161133,
+      "mean_token_accuracy": 0.9407236540317535,
+      "num_tokens": 8864507.0,
+      "step": 3350
+    },
+    {
+      "entropy": 0.26411333337426185,
+      "epoch": 8.333742331288343,
+      "grad_norm": 1.0053174495697021,
+      "learning_rate": 3.082522686786935e-05,
+      "loss": 0.16243309020996094,
+      "mean_token_accuracy": 0.9401588958501815,
+      "num_tokens": 8993933.0,
+      "step": 3400
+    },
+    {
+      "entropy": 0.2589409455657005,
+      "epoch": 8.45644171779141,
+      "grad_norm": 1.1849397420883179,
+      "learning_rate": 2.6571605022016655e-05,
+      "loss": 0.1622679328918457,
+      "mean_token_accuracy": 0.9411727702617645,
+      "num_tokens": 9124819.0,
+      "step": 3450
+    },
+    {
+      "entropy": 0.25455976352095605,
+      "epoch": 8.579141104294479,
+      "grad_norm": 0.9689339995384216,
+      "learning_rate": 2.2611216939830125e-05,
+      "loss": 0.16137266159057617,
+      "mean_token_accuracy": 0.941931157708168,
+      "num_tokens": 9266865.0,
+      "step": 3500
+    },
+    {
+      "entropy": 0.25809177130460736,
+      "epoch": 8.701840490797546,
+      "grad_norm": 1.097219467163086,
+      "learning_rate": 1.8951308748044324e-05,
+      "loss": 0.16276371002197265,
+      "mean_token_accuracy": 0.9418226885795593,
+      "num_tokens": 9402291.0,
+      "step": 3550
+    },
+    {
+      "entropy": 0.2551534166932106,
+      "epoch": 8.824539877300614,
+      "grad_norm": 1.1789699792861938,
+      "learning_rate": 1.5598576800158975e-05,
+      "loss": 0.16089740753173828,
+      "mean_token_accuracy": 0.942240217924118,
+      "num_tokens": 9536676.0,
+      "step": 3600
+    },
+    {
+      "entropy": 0.25731841623783114,
+      "epoch": 8.94723926380368,
+      "grad_norm": 1.071371078491211,
+      "learning_rate": 1.2559155424451429e-05,
+      "loss": 0.16048816680908204,
+      "mean_token_accuracy": 0.9426142275333405,
+      "num_tokens": 9668941.0,
+      "step": 3650
+    },
+    {
+      "epoch": 9.0,
+      "eval_entropy": 0.4000720078604562,
+      "eval_mean_token_accuracy": 0.791837249483381,
+      "eval_not_syn_loss": 0.9112135171890259,
+      "eval_not_syn_runtime": 55.748,
+      "eval_not_syn_samples_per_second": 25.023,
+      "eval_not_syn_steps_per_second": 3.139,
+      "eval_num_tokens": 9721764.0,
+      "step": 3672
+    },
+    {
+      "epoch": 9.0,
+      "eval_entropy": 0.3756853074686868,
+      "eval_mean_token_accuracy": 0.8066899194036211,
+      "eval_num_tokens": 9721764.0,
+      "eval_syn_loss": 0.8509045243263245,
+      "eval_syn_runtime": 56.7641,
+      "eval_syn_samples_per_second": 24.575,
+      "eval_syn_steps_per_second": 3.083,
+      "step": 3672
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 4080,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.932544486512142e+17,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-408/README.md b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-408/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..7ac50e2706e5b4eae8f28e204601afb241c49e55
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-408/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-2B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-2B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-408/adapter_config.json b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-408/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b9596096c71017d49955136a4e23d9f97138d305
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-408/adapter_config.json
@@ -0,0 +1,48 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_bias": false,
+  "lora_dropout": 0.08602048083239339,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "q_proj",
+    "k_proj",
+    "gate_proj",
+    "up_proj",
+    "o_proj",
+    "down_proj",
+    "v_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-408/trainer_state.json b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-408/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..ea01991cd54b8ec069696fc02474103b29397248
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test1/checkpoint-408/trainer_state.json
@@ -0,0 +1,136 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "eval_steps": 500,
+  "global_step": 408,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.2408065378665922,
+      "epoch": 0.12269938650306748,
+      "grad_norm": 1.6718658208847046,
+      "learning_rate": 4.487804970912036e-05,
+      "loss": 2.129624786376953,
+      "mean_token_accuracy": 0.5743644836544991,
+      "num_tokens": 134329.0,
+      "step": 50
+    },
+    {
+      "entropy": 1.1078504550457,
+      "epoch": 0.24539877300613497,
+      "grad_norm": 2.4091813564300537,
+      "learning_rate": 9.067197798373296e-05,
+      "loss": 1.0346390533447265,
+      "mean_token_accuracy": 0.7085381114482879,
+      "num_tokens": 262407.0,
+      "step": 100
+    },
+    {
+      "entropy": 0.9421598988771439,
+      "epoch": 0.36809815950920244,
+      "grad_norm": 1.3913646936416626,
+      "learning_rate": 0.00013646590625834558,
+      "loss": 0.8716551208496094,
+      "mean_token_accuracy": 0.7449040985107422,
+      "num_tokens": 401136.0,
+      "step": 150
+    },
+    {
+      "entropy": 0.8723550146818161,
+      "epoch": 0.49079754601226994,
+      "grad_norm": 1.1664468050003052,
+      "learning_rate": 0.0001822598345329582,
+      "loss": 0.8095977783203125,
+      "mean_token_accuracy": 0.7585240352153778,
+      "num_tokens": 531647.0,
+      "step": 200
+    },
+    {
+      "entropy": 0.8466519421339035,
+      "epoch": 0.6134969325153374,
+      "grad_norm": 1.0145421028137207,
+      "learning_rate": 0.00022805376280757083,
+      "loss": 0.7830857849121093,
+      "mean_token_accuracy": 0.7661705195903779,
+      "num_tokens": 666127.0,
+      "step": 250
+    },
+    {
+      "entropy": 0.8180295366048813,
+      "epoch": 0.7361963190184049,
+      "grad_norm": 0.9731705784797668,
+      "learning_rate": 0.00027384769108218336,
+      "loss": 0.7573603820800782,
+      "mean_token_accuracy": 0.7704453605413437,
+      "num_tokens": 799292.0,
+      "step": 300
+    },
+    {
+      "entropy": 0.7990227991342544,
+      "epoch": 0.8588957055214724,
+      "grad_norm": 1.0642374753952026,
+      "learning_rate": 0.000319641619356796,
+      "loss": 0.73709228515625,
+      "mean_token_accuracy": 0.7752733880281448,
+      "num_tokens": 930614.0,
+      "step": 350
+    },
+    {
+      "entropy": 0.7899087685346603,
+      "epoch": 0.9815950920245399,
+      "grad_norm": 0.880613386631012,
+      "learning_rate": 0.0003654355476314086,
+      "loss": 0.7347319793701171,
+      "mean_token_accuracy": 0.7766731631755829,
+      "num_tokens": 1061654.0,
+      "step": 400
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.83522472347532,
+      "eval_mean_token_accuracy": 0.7484098076820374,
+      "eval_not_syn_loss": 0.7872886061668396,
+      "eval_not_syn_runtime": 54.8697,
+      "eval_not_syn_samples_per_second": 25.424,
+      "eval_not_syn_steps_per_second": 3.189,
+      "eval_num_tokens": 1080196.0,
+      "step": 408
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.7784442271505083,
+      "eval_mean_token_accuracy": 0.8021791185651507,
+      "eval_num_tokens": 1080196.0,
+      "eval_syn_loss": 0.6980098485946655,
+      "eval_syn_runtime": 56.6342,
+      "eval_syn_samples_per_second": 24.632,
+      "eval_syn_steps_per_second": 3.09,
+      "step": 408
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 4080,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.161626494849779e+16,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test2/checkpoint-1632/README.md b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test2/checkpoint-1632/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..7ac50e2706e5b4eae8f28e204601afb241c49e55
--- /dev/null
+++ b/substitutivity_original_Estonian/Qwen3.5-2B-Base_substitutivity_splits_original_features_train_substitutivity_splits_original_features_test2/checkpoint-1632/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-2B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-2B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1371/README.md b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1371/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..7ac50e2706e5b4eae8f28e204601afb241c49e55
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1371/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-2B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-2B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1371/adapter_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1371/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..86d878f780798dd8f06b10d3bdb6eaee702b2da7
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1371/adapter_config.json
@@ -0,0 +1,48 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.01686122009693236,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "v_proj",
+    "o_proj",
+    "k_proj",
+    "up_proj",
+    "q_proj",
+    "down_proj",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1371/chat_template.jinja b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1371/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..0ef09f214eaa6d9bca297988afc1454b5827b2c7
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1371/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is true %}
+        {{- '<think>\n' }}
+    {%- else %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1371/tokenizer_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1371/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1371/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1371/trainer_state.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1371/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..24562f7525284206f24da332305db5343e9a955e
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1371/trainer_state.json
@@ -0,0 +1,337 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.0,
+  "eval_steps": 500,
+  "global_step": 1371,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.074948468208313,
+      "epoch": 0.10952902519167579,
+      "grad_norm": 2.3511903285980225,
+      "learning_rate": 1.2698308676050528e-05,
+      "loss": 1.9153688049316406,
+      "mean_token_accuracy": 0.6077924159169197,
+      "num_tokens": 113087.0,
+      "step": 50
+    },
+    {
+      "entropy": 1.054056368470192,
+      "epoch": 0.21905805038335158,
+      "grad_norm": 2.3081915378570557,
+      "learning_rate": 2.5655766508755147e-05,
+      "loss": 0.9633240509033203,
+      "mean_token_accuracy": 0.741833764910698,
+      "num_tokens": 230392.0,
+      "step": 100
+    },
+    {
+      "entropy": 0.9184817910194397,
+      "epoch": 0.32858707557502737,
+      "grad_norm": 2.0965280532836914,
+      "learning_rate": 3.861322434145977e-05,
+      "loss": 0.8405924987792969,
+      "mean_token_accuracy": 0.7688284432888031,
+      "num_tokens": 341746.0,
+      "step": 150
+    },
+    {
+      "entropy": 0.8618861585855484,
+      "epoch": 0.43811610076670315,
+      "grad_norm": 2.2506542205810547,
+      "learning_rate": 5.157068217416438e-05,
+      "loss": 0.7841030883789063,
+      "mean_token_accuracy": 0.7796466761827469,
+      "num_tokens": 460483.0,
+      "step": 200
+    },
+    {
+      "entropy": 0.819869134426117,
+      "epoch": 0.547645125958379,
+      "grad_norm": 1.4058395624160767,
+      "learning_rate": 6.4528140006869e-05,
+      "loss": 0.7524736785888672,
+      "mean_token_accuracy": 0.7900535291433335,
+      "num_tokens": 570627.0,
+      "step": 250
+    },
+    {
+      "entropy": 0.8075837278366089,
+      "epoch": 0.6571741511500547,
+      "grad_norm": 1.5815014839172363,
+      "learning_rate": 7.748559783957362e-05,
+      "loss": 0.7388697814941406,
+      "mean_token_accuracy": 0.7914055424928665,
+      "num_tokens": 677217.0,
+      "step": 300
+    },
+    {
+      "entropy": 0.7868322026729584,
+      "epoch": 0.7667031763417306,
+      "grad_norm": 1.2377994060516357,
+      "learning_rate": 9.044305567227824e-05,
+      "loss": 0.7226168060302735,
+      "mean_token_accuracy": 0.7953901988267899,
+      "num_tokens": 790185.0,
+      "step": 350
+    },
+    {
+      "entropy": 0.7526325100660324,
+      "epoch": 0.8762322015334063,
+      "grad_norm": 1.203137755393982,
+      "learning_rate": 0.00010340051350498286,
+      "loss": 0.6991680908203125,
+      "mean_token_accuracy": 0.7992895567417144,
+      "num_tokens": 910080.0,
+      "step": 400
+    },
+    {
+      "entropy": 0.7626404595375061,
+      "epoch": 0.9857612267250822,
+      "grad_norm": 1.1625452041625977,
+      "learning_rate": 0.00011635797133768749,
+      "loss": 0.7055020904541016,
+      "mean_token_accuracy": 0.7997505795955658,
+      "num_tokens": 1023473.0,
+      "step": 450
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.7441067624659765,
+      "eval_loss": 0.7209385633468628,
+      "eval_mean_token_accuracy": 0.7895652093584575,
+      "eval_num_tokens": 1038250.0,
+      "eval_runtime": 45.5076,
+      "eval_samples_per_second": 22.018,
+      "eval_steps_per_second": 2.769,
+      "step": 457
+    },
+    {
+      "entropy": 0.720724637460227,
+      "epoch": 1.0941949616648412,
+      "grad_norm": 1.2537180185317993,
+      "learning_rate": 0.00011840069618947894,
+      "loss": 0.6586117553710937,
+      "mean_token_accuracy": 0.8081232917429221,
+      "num_tokens": 1128922.0,
+      "step": 500
+    },
+    {
+      "entropy": 0.7084310132265091,
+      "epoch": 1.203723986856517,
+      "grad_norm": 1.2347161769866943,
+      "learning_rate": 0.00011828501915154198,
+      "loss": 0.650611572265625,
+      "mean_token_accuracy": 0.808387525677681,
+      "num_tokens": 1245635.0,
+      "step": 550
+    },
+    {
+      "entropy": 0.706004958152771,
+      "epoch": 1.3132530120481927,
+      "grad_norm": 1.2140169143676758,
+      "learning_rate": 0.00011808319665683977,
+      "loss": 0.6442465209960937,
+      "mean_token_accuracy": 0.8122780376672745,
+      "num_tokens": 1354885.0,
+      "step": 600
+    },
+    {
+      "entropy": 0.6854291236400605,
+      "epoch": 1.4227820372398685,
+      "grad_norm": 1.1137559413909912,
+      "learning_rate": 0.00011779552303848118,
+      "loss": 0.6293138885498046,
+      "mean_token_accuracy": 0.8158071845769882,
+      "num_tokens": 1472232.0,
+      "step": 650
+    },
+    {
+      "entropy": 0.6885070586204529,
+      "epoch": 1.5323110624315444,
+      "grad_norm": 1.1074714660644531,
+      "learning_rate": 0.00011742241783280468,
+      "loss": 0.6259001541137695,
+      "mean_token_accuracy": 0.8164563828706741,
+      "num_tokens": 1587494.0,
+      "step": 700
+    },
+    {
+      "entropy": 0.6932114177942276,
+      "epoch": 1.6418400876232202,
+      "grad_norm": 1.1099495887756348,
+      "learning_rate": 0.00011696442516753647,
+      "loss": 0.6335408401489258,
+      "mean_token_accuracy": 0.8156683903932571,
+      "num_tokens": 1696927.0,
+      "step": 750
+    },
+    {
+      "entropy": 0.6787053138017655,
+      "epoch": 1.751369112814896,
+      "grad_norm": 1.2505418062210083,
+      "learning_rate": 0.00011642221296824764,
+      "loss": 0.6210909271240235,
+      "mean_token_accuracy": 0.8153067737817764,
+      "num_tokens": 1813415.0,
+      "step": 800
+    },
+    {
+      "entropy": 0.6693948286771775,
+      "epoch": 1.8608981380065717,
+      "grad_norm": 1.0963624715805054,
+      "learning_rate": 0.00011579657198426736,
+      "loss": 0.6105814361572266,
+      "mean_token_accuracy": 0.819609425663948,
+      "num_tokens": 1924911.0,
+      "step": 850
+    },
+    {
+      "entropy": 0.6634757363796234,
+      "epoch": 1.9704271631982475,
+      "grad_norm": 0.8766036033630371,
+      "learning_rate": 0.00011508841463547313,
+      "loss": 0.6100498580932617,
+      "mean_token_accuracy": 0.8197187691926956,
+      "num_tokens": 2044650.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.0,
+      "eval_entropy": 0.6342300275961558,
+      "eval_loss": 0.6517631411552429,
+      "eval_mean_token_accuracy": 0.807878240233376,
+      "eval_num_tokens": 2076500.0,
+      "eval_runtime": 44.6225,
+      "eval_samples_per_second": 22.455,
+      "eval_steps_per_second": 2.824,
+      "step": 914
+    },
+    {
+      "entropy": 0.6060671378867795,
+      "epoch": 2.078860898138007,
+      "grad_norm": 0.8028743267059326,
+      "learning_rate": 0.00011429877368163915,
+      "loss": 0.5440861511230469,
+      "mean_token_accuracy": 0.833196161973356,
+      "num_tokens": 2160498.0,
+      "step": 950
+    },
+    {
+      "entropy": 0.5839239662885666,
+      "epoch": 2.1883899233296824,
+      "grad_norm": 0.8585777878761292,
+      "learning_rate": 0.00011342880071628435,
+      "loss": 0.5182851409912109,
+      "mean_token_accuracy": 0.8406627786159515,
+      "num_tokens": 2269979.0,
+      "step": 1000
+    },
+    {
+      "entropy": 0.5949062475562096,
+      "epoch": 2.297918948521358,
+      "grad_norm": 0.9953155517578125,
+      "learning_rate": 0.00011247976448721578,
+      "loss": 0.5327045822143555,
+      "mean_token_accuracy": 0.8344444561004639,
+      "num_tokens": 2378362.0,
+      "step": 1050
+    },
+    {
+      "entropy": 0.583380132317543,
+      "epoch": 2.407447973713034,
+      "grad_norm": 0.9235143065452576,
+      "learning_rate": 0.00011145304904621704,
+      "loss": 0.520841178894043,
+      "mean_token_accuracy": 0.839934149980545,
+      "num_tokens": 2498494.0,
+      "step": 1100
+    },
+    {
+      "entropy": 0.5811308068037033,
+      "epoch": 2.5169769989047097,
+      "grad_norm": 1.1114075183868408,
+      "learning_rate": 0.00011035015173058038,
+      "loss": 0.5212027740478515,
+      "mean_token_accuracy": 0.838929146528244,
+      "num_tokens": 2611646.0,
+      "step": 1150
+    },
+    {
+      "entropy": 0.5903497302532196,
+      "epoch": 2.6265060240963853,
+      "grad_norm": 1.053895354270935,
+      "learning_rate": 0.0001091726809794255,
+      "loss": 0.5264106750488281,
+      "mean_token_accuracy": 0.8369076484441758,
+      "num_tokens": 2717744.0,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5949014449119567,
+      "epoch": 2.7360350492880614,
+      "grad_norm": 0.7563770413398743,
+      "learning_rate": 0.00010792235398799048,
+      "loss": 0.5322903442382813,
+      "mean_token_accuracy": 0.8368168956041336,
+      "num_tokens": 2827750.0,
+      "step": 1250
+    },
+    {
+      "entropy": 0.5745143675804139,
+      "epoch": 2.845564074479737,
+      "grad_norm": 0.8103823065757751,
+      "learning_rate": 0.00010660099420331526,
+      "loss": 0.5154821014404297,
+      "mean_token_accuracy": 0.8410378849506378,
+      "num_tokens": 2948958.0,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5665904894471169,
+      "epoch": 2.955093099671413,
+      "grad_norm": 0.8764206171035767,
+      "learning_rate": 0.00010521052866496986,
+      "loss": 0.5121800231933594,
+      "mean_token_accuracy": 0.8420848768949508,
+      "num_tokens": 3066917.0,
+      "step": 1350
+    },
+    {
+      "epoch": 3.0,
+      "eval_entropy": 0.5603872095308606,
+      "eval_loss": 0.6720245480537415,
+      "eval_mean_token_accuracy": 0.8079002830717299,
+      "eval_num_tokens": 3114750.0,
+      "eval_runtime": 43.9396,
+      "eval_samples_per_second": 22.804,
+      "eval_steps_per_second": 2.868,
+      "step": 1371
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 4570,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 6.747546481924416e+16,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1828/README.md b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1828/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..7ac50e2706e5b4eae8f28e204601afb241c49e55
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1828/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-2B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-2B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1828/adapter_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1828/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..86d878f780798dd8f06b10d3bdb6eaee702b2da7
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1828/adapter_config.json
@@ -0,0 +1,48 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.01686122009693236,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "v_proj",
+    "o_proj",
+    "k_proj",
+    "up_proj",
+    "q_proj",
+    "down_proj",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1828/chat_template.jinja b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1828/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..0ef09f214eaa6d9bca297988afc1454b5827b2c7
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1828/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is true %}
+        {{- '<think>\n' }}
+    {%- else %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1828/tokenizer_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1828/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1828/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1828/trainer_state.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1828/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..4386097ac97c81cf9541e3609e65aaa557db1ae5
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-1828/trainer_state.json
@@ -0,0 +1,438 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 4.0,
+  "eval_steps": 500,
+  "global_step": 1828,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.074948468208313,
+      "epoch": 0.10952902519167579,
+      "grad_norm": 2.3511903285980225,
+      "learning_rate": 1.2698308676050528e-05,
+      "loss": 1.9153688049316406,
+      "mean_token_accuracy": 0.6077924159169197,
+      "num_tokens": 113087.0,
+      "step": 50
+    },
+    {
+      "entropy": 1.054056368470192,
+      "epoch": 0.21905805038335158,
+      "grad_norm": 2.3081915378570557,
+      "learning_rate": 2.5655766508755147e-05,
+      "loss": 0.9633240509033203,
+      "mean_token_accuracy": 0.741833764910698,
+      "num_tokens": 230392.0,
+      "step": 100
+    },
+    {
+      "entropy": 0.9184817910194397,
+      "epoch": 0.32858707557502737,
+      "grad_norm": 2.0965280532836914,
+      "learning_rate": 3.861322434145977e-05,
+      "loss": 0.8405924987792969,
+      "mean_token_accuracy": 0.7688284432888031,
+      "num_tokens": 341746.0,
+      "step": 150
+    },
+    {
+      "entropy": 0.8618861585855484,
+      "epoch": 0.43811610076670315,
+      "grad_norm": 2.2506542205810547,
+      "learning_rate": 5.157068217416438e-05,
+      "loss": 0.7841030883789063,
+      "mean_token_accuracy": 0.7796466761827469,
+      "num_tokens": 460483.0,
+      "step": 200
+    },
+    {
+      "entropy": 0.819869134426117,
+      "epoch": 0.547645125958379,
+      "grad_norm": 1.4058395624160767,
+      "learning_rate": 6.4528140006869e-05,
+      "loss": 0.7524736785888672,
+      "mean_token_accuracy": 0.7900535291433335,
+      "num_tokens": 570627.0,
+      "step": 250
+    },
+    {
+      "entropy": 0.8075837278366089,
+      "epoch": 0.6571741511500547,
+      "grad_norm": 1.5815014839172363,
+      "learning_rate": 7.748559783957362e-05,
+      "loss": 0.7388697814941406,
+      "mean_token_accuracy": 0.7914055424928665,
+      "num_tokens": 677217.0,
+      "step": 300
+    },
+    {
+      "entropy": 0.7868322026729584,
+      "epoch": 0.7667031763417306,
+      "grad_norm": 1.2377994060516357,
+      "learning_rate": 9.044305567227824e-05,
+      "loss": 0.7226168060302735,
+      "mean_token_accuracy": 0.7953901988267899,
+      "num_tokens": 790185.0,
+      "step": 350
+    },
+    {
+      "entropy": 0.7526325100660324,
+      "epoch": 0.8762322015334063,
+      "grad_norm": 1.203137755393982,
+      "learning_rate": 0.00010340051350498286,
+      "loss": 0.6991680908203125,
+      "mean_token_accuracy": 0.7992895567417144,
+      "num_tokens": 910080.0,
+      "step": 400
+    },
+    {
+      "entropy": 0.7626404595375061,
+      "epoch": 0.9857612267250822,
+      "grad_norm": 1.1625452041625977,
+      "learning_rate": 0.00011635797133768749,
+      "loss": 0.7055020904541016,
+      "mean_token_accuracy": 0.7997505795955658,
+      "num_tokens": 1023473.0,
+      "step": 450
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.7441067624659765,
+      "eval_loss": 0.7209385633468628,
+      "eval_mean_token_accuracy": 0.7895652093584575,
+      "eval_num_tokens": 1038250.0,
+      "eval_runtime": 45.5076,
+      "eval_samples_per_second": 22.018,
+      "eval_steps_per_second": 2.769,
+      "step": 457
+    },
+    {
+      "entropy": 0.720724637460227,
+      "epoch": 1.0941949616648412,
+      "grad_norm": 1.2537180185317993,
+      "learning_rate": 0.00011840069618947894,
+      "loss": 0.6586117553710937,
+      "mean_token_accuracy": 0.8081232917429221,
+      "num_tokens": 1128922.0,
+      "step": 500
+    },
+    {
+      "entropy": 0.7084310132265091,
+      "epoch": 1.203723986856517,
+      "grad_norm": 1.2347161769866943,
+      "learning_rate": 0.00011828501915154198,
+      "loss": 0.650611572265625,
+      "mean_token_accuracy": 0.808387525677681,
+      "num_tokens": 1245635.0,
+      "step": 550
+    },
+    {
+      "entropy": 0.706004958152771,
+      "epoch": 1.3132530120481927,
+      "grad_norm": 1.2140169143676758,
+      "learning_rate": 0.00011808319665683977,
+      "loss": 0.6442465209960937,
+      "mean_token_accuracy": 0.8122780376672745,
+      "num_tokens": 1354885.0,
+      "step": 600
+    },
+    {
+      "entropy": 0.6854291236400605,
+      "epoch": 1.4227820372398685,
+      "grad_norm": 1.1137559413909912,
+      "learning_rate": 0.00011779552303848118,
+      "loss": 0.6293138885498046,
+      "mean_token_accuracy": 0.8158071845769882,
+      "num_tokens": 1472232.0,
+      "step": 650
+    },
+    {
+      "entropy": 0.6885070586204529,
+      "epoch": 1.5323110624315444,
+      "grad_norm": 1.1074714660644531,
+      "learning_rate": 0.00011742241783280468,
+      "loss": 0.6259001541137695,
+      "mean_token_accuracy": 0.8164563828706741,
+      "num_tokens": 1587494.0,
+      "step": 700
+    },
+    {
+      "entropy": 0.6932114177942276,
+      "epoch": 1.6418400876232202,
+      "grad_norm": 1.1099495887756348,
+      "learning_rate": 0.00011696442516753647,
+      "loss": 0.6335408401489258,
+      "mean_token_accuracy": 0.8156683903932571,
+      "num_tokens": 1696927.0,
+      "step": 750
+    },
+    {
+      "entropy": 0.6787053138017655,
+      "epoch": 1.751369112814896,
+      "grad_norm": 1.2505418062210083,
+      "learning_rate": 0.00011642221296824764,
+      "loss": 0.6210909271240235,
+      "mean_token_accuracy": 0.8153067737817764,
+      "num_tokens": 1813415.0,
+      "step": 800
+    },
+    {
+      "entropy": 0.6693948286771775,
+      "epoch": 1.8608981380065717,
+      "grad_norm": 1.0963624715805054,
+      "learning_rate": 0.00011579657198426736,
+      "loss": 0.6105814361572266,
+      "mean_token_accuracy": 0.819609425663948,
+      "num_tokens": 1924911.0,
+      "step": 850
+    },
+    {
+      "entropy": 0.6634757363796234,
+      "epoch": 1.9704271631982475,
+      "grad_norm": 0.8766036033630371,
+      "learning_rate": 0.00011508841463547313,
+      "loss": 0.6100498580932617,
+      "mean_token_accuracy": 0.8197187691926956,
+      "num_tokens": 2044650.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.0,
+      "eval_entropy": 0.6342300275961558,
+      "eval_loss": 0.6517631411552429,
+      "eval_mean_token_accuracy": 0.807878240233376,
+      "eval_num_tokens": 2076500.0,
+      "eval_runtime": 44.6225,
+      "eval_samples_per_second": 22.455,
+      "eval_steps_per_second": 2.824,
+      "step": 914
+    },
+    {
+      "entropy": 0.6060671378867795,
+      "epoch": 2.078860898138007,
+      "grad_norm": 0.8028743267059326,
+      "learning_rate": 0.00011429877368163915,
+      "loss": 0.5440861511230469,
+      "mean_token_accuracy": 0.833196161973356,
+      "num_tokens": 2160498.0,
+      "step": 950
+    },
+    {
+      "entropy": 0.5839239662885666,
+      "epoch": 2.1883899233296824,
+      "grad_norm": 0.8585777878761292,
+      "learning_rate": 0.00011342880071628435,
+      "loss": 0.5182851409912109,
+      "mean_token_accuracy": 0.8406627786159515,
+      "num_tokens": 2269979.0,
+      "step": 1000
+    },
+    {
+      "entropy": 0.5949062475562096,
+      "epoch": 2.297918948521358,
+      "grad_norm": 0.9953155517578125,
+      "learning_rate": 0.00011247976448721578,
+      "loss": 0.5327045822143555,
+      "mean_token_accuracy": 0.8344444561004639,
+      "num_tokens": 2378362.0,
+      "step": 1050
+    },
+    {
+      "entropy": 0.583380132317543,
+      "epoch": 2.407447973713034,
+      "grad_norm": 0.9235143065452576,
+      "learning_rate": 0.00011145304904621704,
+      "loss": 0.520841178894043,
+      "mean_token_accuracy": 0.839934149980545,
+      "num_tokens": 2498494.0,
+      "step": 1100
+    },
+    {
+      "entropy": 0.5811308068037033,
+      "epoch": 2.5169769989047097,
+      "grad_norm": 1.1114075183868408,
+      "learning_rate": 0.00011035015173058038,
+      "loss": 0.5212027740478515,
+      "mean_token_accuracy": 0.838929146528244,
+      "num_tokens": 2611646.0,
+      "step": 1150
+    },
+    {
+      "entropy": 0.5903497302532196,
+      "epoch": 2.6265060240963853,
+      "grad_norm": 1.053895354270935,
+      "learning_rate": 0.0001091726809794255,
+      "loss": 0.5264106750488281,
+      "mean_token_accuracy": 0.8369076484441758,
+      "num_tokens": 2717744.0,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5949014449119567,
+      "epoch": 2.7360350492880614,
+      "grad_norm": 0.7563770413398743,
+      "learning_rate": 0.00010792235398799048,
+      "loss": 0.5322903442382813,
+      "mean_token_accuracy": 0.8368168956041336,
+      "num_tokens": 2827750.0,
+      "step": 1250
+    },
+    {
+      "entropy": 0.5745143675804139,
+      "epoch": 2.845564074479737,
+      "grad_norm": 0.8103823065757751,
+      "learning_rate": 0.00010660099420331526,
+      "loss": 0.5154821014404297,
+      "mean_token_accuracy": 0.8410378849506378,
+      "num_tokens": 2948958.0,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5665904894471169,
+      "epoch": 2.955093099671413,
+      "grad_norm": 0.8764206171035767,
+      "learning_rate": 0.00010521052866496986,
+      "loss": 0.5121800231933594,
+      "mean_token_accuracy": 0.8420848768949508,
+      "num_tokens": 3066917.0,
+      "step": 1350
+    },
+    {
+      "epoch": 3.0,
+      "eval_entropy": 0.5603872095308606,
+      "eval_loss": 0.6720245480537415,
+      "eval_mean_token_accuracy": 0.8079002830717299,
+      "eval_num_tokens": 3114750.0,
+      "eval_runtime": 43.9396,
+      "eval_samples_per_second": 22.804,
+      "eval_steps_per_second": 2.868,
+      "step": 1371
+    },
+    {
+      "entropy": 0.5181072778774031,
+      "epoch": 3.063526834611172,
+      "grad_norm": 0.889443039894104,
+      "learning_rate": 0.00010375298519470596,
+      "loss": 0.44869777679443357,
+      "mean_token_accuracy": 0.8574847211741438,
+      "num_tokens": 3180366.0,
+      "step": 1400
+    },
+    {
+      "entropy": 0.4902029836177826,
+      "epoch": 3.1730558598028478,
+      "grad_norm": 0.8999125957489014,
+      "learning_rate": 0.00010223048943913007,
+      "loss": 0.41646446228027345,
+      "mean_token_accuracy": 0.8655553787946701,
+      "num_tokens": 3293918.0,
+      "step": 1450
+    },
+    {
+      "entropy": 0.48207574665546415,
+      "epoch": 3.2825848849945234,
+      "grad_norm": 1.0042189359664917,
+      "learning_rate": 0.00010064526176971124,
+      "loss": 0.41205951690673825,
+      "mean_token_accuracy": 0.866786903142929,
+      "num_tokens": 3409058.0,
+      "step": 1500
+    },
+    {
+      "entropy": 0.4839185461401939,
+      "epoch": 3.3921139101861995,
+      "grad_norm": 1.1650396585464478,
+      "learning_rate": 9.89996140446443e-05,
+      "loss": 0.4102657699584961,
+      "mean_token_accuracy": 0.8668668657541275,
+      "num_tokens": 3525131.0,
+      "step": 1550
+    },
+    {
+      "entropy": 0.4932769918441772,
+      "epoch": 3.501642935377875,
+      "grad_norm": 1.0935224294662476,
+      "learning_rate": 9.729594623729099e-05,
+      "loss": 0.4225375366210937,
+      "mean_token_accuracy": 0.8634544950723648,
+      "num_tokens": 3641504.0,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5046219238638878,
+      "epoch": 3.6111719605695507,
+      "grad_norm": 0.9509316086769104,
+      "learning_rate": 9.553674293611626e-05,
+      "loss": 0.4319529342651367,
+      "mean_token_accuracy": 0.8610132175683975,
+      "num_tokens": 3752182.0,
+      "step": 1650
+    },
+    {
+      "entropy": 0.5090703725814819,
+      "epoch": 3.7207009857612268,
+      "grad_norm": 0.9050212502479553,
+      "learning_rate": 9.372456972122359e-05,
+      "loss": 0.4376397705078125,
+      "mean_token_accuracy": 0.8591927194595337,
+      "num_tokens": 3866473.0,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4903561845421791,
+      "epoch": 3.8302300109529024,
+      "grad_norm": 1.1624391078948975,
+      "learning_rate": 9.186206942277429e-05,
+      "loss": 0.4219248962402344,
+      "mean_token_accuracy": 0.8636374253034592,
+      "num_tokens": 3977393.0,
+      "step": 1750
+    },
+    {
+      "entropy": 0.5040662395954132,
+      "epoch": 3.9397590361445785,
+      "grad_norm": 0.798085629940033,
+      "learning_rate": 8.995195826674703e-05,
+      "loss": 0.4333121871948242,
+      "mean_token_accuracy": 0.86187963783741,
+      "num_tokens": 4088936.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5136857172326436,
+      "eval_loss": 0.6814945340156555,
+      "eval_mean_token_accuracy": 0.8091600262929523,
+      "eval_num_tokens": 4153000.0,
+      "eval_runtime": 44.5059,
+      "eval_samples_per_second": 22.514,
+      "eval_steps_per_second": 2.831,
+      "step": 1828
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 4570,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 8.985192850101312e+16,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2285/README.md b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2285/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..7ac50e2706e5b4eae8f28e204601afb241c49e55
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2285/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-2B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-2B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2285/adapter_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2285/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..86d878f780798dd8f06b10d3bdb6eaee702b2da7
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2285/adapter_config.json
@@ -0,0 +1,48 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.01686122009693236,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "v_proj",
+    "o_proj",
+    "k_proj",
+    "up_proj",
+    "q_proj",
+    "down_proj",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2285/chat_template.jinja b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2285/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..0ef09f214eaa6d9bca297988afc1454b5827b2c7
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2285/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is true %}
+        {{- '<think>\n' }}
+    {%- else %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2285/tokenizer_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2285/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2285/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2285/trainer_state.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2285/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..89931eb1a046b8f31066e8dfad6284df94ebcc0f
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2285/trainer_state.json
@@ -0,0 +1,539 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 5.0,
+  "eval_steps": 500,
+  "global_step": 2285,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.074948468208313,
+      "epoch": 0.10952902519167579,
+      "grad_norm": 2.3511903285980225,
+      "learning_rate": 1.2698308676050528e-05,
+      "loss": 1.9153688049316406,
+      "mean_token_accuracy": 0.6077924159169197,
+      "num_tokens": 113087.0,
+      "step": 50
+    },
+    {
+      "entropy": 1.054056368470192,
+      "epoch": 0.21905805038335158,
+      "grad_norm": 2.3081915378570557,
+      "learning_rate": 2.5655766508755147e-05,
+      "loss": 0.9633240509033203,
+      "mean_token_accuracy": 0.741833764910698,
+      "num_tokens": 230392.0,
+      "step": 100
+    },
+    {
+      "entropy": 0.9184817910194397,
+      "epoch": 0.32858707557502737,
+      "grad_norm": 2.0965280532836914,
+      "learning_rate": 3.861322434145977e-05,
+      "loss": 0.8405924987792969,
+      "mean_token_accuracy": 0.7688284432888031,
+      "num_tokens": 341746.0,
+      "step": 150
+    },
+    {
+      "entropy": 0.8618861585855484,
+      "epoch": 0.43811610076670315,
+      "grad_norm": 2.2506542205810547,
+      "learning_rate": 5.157068217416438e-05,
+      "loss": 0.7841030883789063,
+      "mean_token_accuracy": 0.7796466761827469,
+      "num_tokens": 460483.0,
+      "step": 200
+    },
+    {
+      "entropy": 0.819869134426117,
+      "epoch": 0.547645125958379,
+      "grad_norm": 1.4058395624160767,
+      "learning_rate": 6.4528140006869e-05,
+      "loss": 0.7524736785888672,
+      "mean_token_accuracy": 0.7900535291433335,
+      "num_tokens": 570627.0,
+      "step": 250
+    },
+    {
+      "entropy": 0.8075837278366089,
+      "epoch": 0.6571741511500547,
+      "grad_norm": 1.5815014839172363,
+      "learning_rate": 7.748559783957362e-05,
+      "loss": 0.7388697814941406,
+      "mean_token_accuracy": 0.7914055424928665,
+      "num_tokens": 677217.0,
+      "step": 300
+    },
+    {
+      "entropy": 0.7868322026729584,
+      "epoch": 0.7667031763417306,
+      "grad_norm": 1.2377994060516357,
+      "learning_rate": 9.044305567227824e-05,
+      "loss": 0.7226168060302735,
+      "mean_token_accuracy": 0.7953901988267899,
+      "num_tokens": 790185.0,
+      "step": 350
+    },
+    {
+      "entropy": 0.7526325100660324,
+      "epoch": 0.8762322015334063,
+      "grad_norm": 1.203137755393982,
+      "learning_rate": 0.00010340051350498286,
+      "loss": 0.6991680908203125,
+      "mean_token_accuracy": 0.7992895567417144,
+      "num_tokens": 910080.0,
+      "step": 400
+    },
+    {
+      "entropy": 0.7626404595375061,
+      "epoch": 0.9857612267250822,
+      "grad_norm": 1.1625452041625977,
+      "learning_rate": 0.00011635797133768749,
+      "loss": 0.7055020904541016,
+      "mean_token_accuracy": 0.7997505795955658,
+      "num_tokens": 1023473.0,
+      "step": 450
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.7441067624659765,
+      "eval_loss": 0.7209385633468628,
+      "eval_mean_token_accuracy": 0.7895652093584575,
+      "eval_num_tokens": 1038250.0,
+      "eval_runtime": 45.5076,
+      "eval_samples_per_second": 22.018,
+      "eval_steps_per_second": 2.769,
+      "step": 457
+    },
+    {
+      "entropy": 0.720724637460227,
+      "epoch": 1.0941949616648412,
+      "grad_norm": 1.2537180185317993,
+      "learning_rate": 0.00011840069618947894,
+      "loss": 0.6586117553710937,
+      "mean_token_accuracy": 0.8081232917429221,
+      "num_tokens": 1128922.0,
+      "step": 500
+    },
+    {
+      "entropy": 0.7084310132265091,
+      "epoch": 1.203723986856517,
+      "grad_norm": 1.2347161769866943,
+      "learning_rate": 0.00011828501915154198,
+      "loss": 0.650611572265625,
+      "mean_token_accuracy": 0.808387525677681,
+      "num_tokens": 1245635.0,
+      "step": 550
+    },
+    {
+      "entropy": 0.706004958152771,
+      "epoch": 1.3132530120481927,
+      "grad_norm": 1.2140169143676758,
+      "learning_rate": 0.00011808319665683977,
+      "loss": 0.6442465209960937,
+      "mean_token_accuracy": 0.8122780376672745,
+      "num_tokens": 1354885.0,
+      "step": 600
+    },
+    {
+      "entropy": 0.6854291236400605,
+      "epoch": 1.4227820372398685,
+      "grad_norm": 1.1137559413909912,
+      "learning_rate": 0.00011779552303848118,
+      "loss": 0.6293138885498046,
+      "mean_token_accuracy": 0.8158071845769882,
+      "num_tokens": 1472232.0,
+      "step": 650
+    },
+    {
+      "entropy": 0.6885070586204529,
+      "epoch": 1.5323110624315444,
+      "grad_norm": 1.1074714660644531,
+      "learning_rate": 0.00011742241783280468,
+      "loss": 0.6259001541137695,
+      "mean_token_accuracy": 0.8164563828706741,
+      "num_tokens": 1587494.0,
+      "step": 700
+    },
+    {
+      "entropy": 0.6932114177942276,
+      "epoch": 1.6418400876232202,
+      "grad_norm": 1.1099495887756348,
+      "learning_rate": 0.00011696442516753647,
+      "loss": 0.6335408401489258,
+      "mean_token_accuracy": 0.8156683903932571,
+      "num_tokens": 1696927.0,
+      "step": 750
+    },
+    {
+      "entropy": 0.6787053138017655,
+      "epoch": 1.751369112814896,
+      "grad_norm": 1.2505418062210083,
+      "learning_rate": 0.00011642221296824764,
+      "loss": 0.6210909271240235,
+      "mean_token_accuracy": 0.8153067737817764,
+      "num_tokens": 1813415.0,
+      "step": 800
+    },
+    {
+      "entropy": 0.6693948286771775,
+      "epoch": 1.8608981380065717,
+      "grad_norm": 1.0963624715805054,
+      "learning_rate": 0.00011579657198426736,
+      "loss": 0.6105814361572266,
+      "mean_token_accuracy": 0.819609425663948,
+      "num_tokens": 1924911.0,
+      "step": 850
+    },
+    {
+      "entropy": 0.6634757363796234,
+      "epoch": 1.9704271631982475,
+      "grad_norm": 0.8766036033630371,
+      "learning_rate": 0.00011508841463547313,
+      "loss": 0.6100498580932617,
+      "mean_token_accuracy": 0.8197187691926956,
+      "num_tokens": 2044650.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.0,
+      "eval_entropy": 0.6342300275961558,
+      "eval_loss": 0.6517631411552429,
+      "eval_mean_token_accuracy": 0.807878240233376,
+      "eval_num_tokens": 2076500.0,
+      "eval_runtime": 44.6225,
+      "eval_samples_per_second": 22.455,
+      "eval_steps_per_second": 2.824,
+      "step": 914
+    },
+    {
+      "entropy": 0.6060671378867795,
+      "epoch": 2.078860898138007,
+      "grad_norm": 0.8028743267059326,
+      "learning_rate": 0.00011429877368163915,
+      "loss": 0.5440861511230469,
+      "mean_token_accuracy": 0.833196161973356,
+      "num_tokens": 2160498.0,
+      "step": 950
+    },
+    {
+      "entropy": 0.5839239662885666,
+      "epoch": 2.1883899233296824,
+      "grad_norm": 0.8585777878761292,
+      "learning_rate": 0.00011342880071628435,
+      "loss": 0.5182851409912109,
+      "mean_token_accuracy": 0.8406627786159515,
+      "num_tokens": 2269979.0,
+      "step": 1000
+    },
+    {
+      "entropy": 0.5949062475562096,
+      "epoch": 2.297918948521358,
+      "grad_norm": 0.9953155517578125,
+      "learning_rate": 0.00011247976448721578,
+      "loss": 0.5327045822143555,
+      "mean_token_accuracy": 0.8344444561004639,
+      "num_tokens": 2378362.0,
+      "step": 1050
+    },
+    {
+      "entropy": 0.583380132317543,
+      "epoch": 2.407447973713034,
+      "grad_norm": 0.9235143065452576,
+      "learning_rate": 0.00011145304904621704,
+      "loss": 0.520841178894043,
+      "mean_token_accuracy": 0.839934149980545,
+      "num_tokens": 2498494.0,
+      "step": 1100
+    },
+    {
+      "entropy": 0.5811308068037033,
+      "epoch": 2.5169769989047097,
+      "grad_norm": 1.1114075183868408,
+      "learning_rate": 0.00011035015173058038,
+      "loss": 0.5212027740478515,
+      "mean_token_accuracy": 0.838929146528244,
+      "num_tokens": 2611646.0,
+      "step": 1150
+    },
+    {
+      "entropy": 0.5903497302532196,
+      "epoch": 2.6265060240963853,
+      "grad_norm": 1.053895354270935,
+      "learning_rate": 0.0001091726809794255,
+      "loss": 0.5264106750488281,
+      "mean_token_accuracy": 0.8369076484441758,
+      "num_tokens": 2717744.0,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5949014449119567,
+      "epoch": 2.7360350492880614,
+      "grad_norm": 0.7563770413398743,
+      "learning_rate": 0.00010792235398799048,
+      "loss": 0.5322903442382813,
+      "mean_token_accuracy": 0.8368168956041336,
+      "num_tokens": 2827750.0,
+      "step": 1250
+    },
+    {
+      "entropy": 0.5745143675804139,
+      "epoch": 2.845564074479737,
+      "grad_norm": 0.8103823065757751,
+      "learning_rate": 0.00010660099420331526,
+      "loss": 0.5154821014404297,
+      "mean_token_accuracy": 0.8410378849506378,
+      "num_tokens": 2948958.0,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5665904894471169,
+      "epoch": 2.955093099671413,
+      "grad_norm": 0.8764206171035767,
+      "learning_rate": 0.00010521052866496986,
+      "loss": 0.5121800231933594,
+      "mean_token_accuracy": 0.8420848768949508,
+      "num_tokens": 3066917.0,
+      "step": 1350
+    },
+    {
+      "epoch": 3.0,
+      "eval_entropy": 0.5603872095308606,
+      "eval_loss": 0.6720245480537415,
+      "eval_mean_token_accuracy": 0.8079002830717299,
+      "eval_num_tokens": 3114750.0,
+      "eval_runtime": 43.9396,
+      "eval_samples_per_second": 22.804,
+      "eval_steps_per_second": 2.868,
+      "step": 1371
+    },
+    {
+      "entropy": 0.5181072778774031,
+      "epoch": 3.063526834611172,
+      "grad_norm": 0.889443039894104,
+      "learning_rate": 0.00010375298519470596,
+      "loss": 0.44869777679443357,
+      "mean_token_accuracy": 0.8574847211741438,
+      "num_tokens": 3180366.0,
+      "step": 1400
+    },
+    {
+      "entropy": 0.4902029836177826,
+      "epoch": 3.1730558598028478,
+      "grad_norm": 0.8999125957489014,
+      "learning_rate": 0.00010223048943913007,
+      "loss": 0.41646446228027345,
+      "mean_token_accuracy": 0.8655553787946701,
+      "num_tokens": 3293918.0,
+      "step": 1450
+    },
+    {
+      "entropy": 0.48207574665546415,
+      "epoch": 3.2825848849945234,
+      "grad_norm": 1.0042189359664917,
+      "learning_rate": 0.00010064526176971124,
+      "loss": 0.41205951690673825,
+      "mean_token_accuracy": 0.866786903142929,
+      "num_tokens": 3409058.0,
+      "step": 1500
+    },
+    {
+      "entropy": 0.4839185461401939,
+      "epoch": 3.3921139101861995,
+      "grad_norm": 1.1650396585464478,
+      "learning_rate": 9.89996140446443e-05,
+      "loss": 0.4102657699584961,
+      "mean_token_accuracy": 0.8668668657541275,
+      "num_tokens": 3525131.0,
+      "step": 1550
+    },
+    {
+      "entropy": 0.4932769918441772,
+      "epoch": 3.501642935377875,
+      "grad_norm": 1.0935224294662476,
+      "learning_rate": 9.729594623729099e-05,
+      "loss": 0.4225375366210937,
+      "mean_token_accuracy": 0.8634544950723648,
+      "num_tokens": 3641504.0,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5046219238638878,
+      "epoch": 3.6111719605695507,
+      "grad_norm": 0.9509316086769104,
+      "learning_rate": 9.553674293611626e-05,
+      "loss": 0.4319529342651367,
+      "mean_token_accuracy": 0.8610132175683975,
+      "num_tokens": 3752182.0,
+      "step": 1650
+    },
+    {
+      "entropy": 0.5090703725814819,
+      "epoch": 3.7207009857612268,
+      "grad_norm": 0.9050212502479553,
+      "learning_rate": 9.372456972122359e-05,
+      "loss": 0.4376397705078125,
+      "mean_token_accuracy": 0.8591927194595337,
+      "num_tokens": 3866473.0,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4903561845421791,
+      "epoch": 3.8302300109529024,
+      "grad_norm": 1.1624391078948975,
+      "learning_rate": 9.186206942277429e-05,
+      "loss": 0.4219248962402344,
+      "mean_token_accuracy": 0.8636374253034592,
+      "num_tokens": 3977393.0,
+      "step": 1750
+    },
+    {
+      "entropy": 0.5040662395954132,
+      "epoch": 3.9397590361445785,
+      "grad_norm": 0.798085629940033,
+      "learning_rate": 8.995195826674703e-05,
+      "loss": 0.4333121871948242,
+      "mean_token_accuracy": 0.86187963783741,
+      "num_tokens": 4088936.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5136857172326436,
+      "eval_loss": 0.6814945340156555,
+      "eval_mean_token_accuracy": 0.8091600262929523,
+      "eval_num_tokens": 4153000.0,
+      "eval_runtime": 44.5059,
+      "eval_samples_per_second": 22.514,
+      "eval_steps_per_second": 2.831,
+      "step": 1828
+    },
+    {
+      "entropy": 0.4410170723091472,
+      "epoch": 4.048192771084337,
+      "grad_norm": 0.9169163703918457,
+      "learning_rate": 8.799702191365879e-05,
+      "loss": 0.3653216552734375,
+      "mean_token_accuracy": 0.8799899684058295,
+      "num_tokens": 4202329.0,
+      "step": 1850
+    },
+    {
+      "entropy": 0.38991558194160464,
+      "epoch": 4.157721796276014,
+      "grad_norm": 1.2269299030303955,
+      "learning_rate": 8.600011139602396e-05,
+      "loss": 0.3035482597351074,
+      "mean_token_accuracy": 0.8981181675195694,
+      "num_tokens": 4320321.0,
+      "step": 1900
+    },
+    {
+      "entropy": 0.38429499566555025,
+      "epoch": 4.267250821467689,
+      "grad_norm": 1.1254630088806152,
+      "learning_rate": 8.39641389604766e-05,
+      "loss": 0.3037366676330566,
+      "mean_token_accuracy": 0.8979371869564057,
+      "num_tokens": 4435139.0,
+      "step": 1950
+    },
+    {
+      "entropy": 0.39086049795150757,
+      "epoch": 4.376779846659365,
+      "grad_norm": 1.3771028518676758,
+      "learning_rate": 8.18920738206195e-05,
+      "loss": 0.3100498390197754,
+      "mean_token_accuracy": 0.8953604429960251,
+      "num_tokens": 4547272.0,
+      "step": 2000
+    },
+    {
+      "entropy": 0.3967577290534973,
+      "epoch": 4.48630887185104,
+      "grad_norm": 1.0673140287399292,
+      "learning_rate": 7.978693782679394e-05,
+      "loss": 0.31265506744384763,
+      "mean_token_accuracy": 0.8946040958166123,
+      "num_tokens": 4659036.0,
+      "step": 2050
+    },
+    {
+      "entropy": 0.3927252873778343,
+      "epoch": 4.595837897042716,
+      "grad_norm": 1.1611689329147339,
+      "learning_rate": 7.765180105908515e-05,
+      "loss": 0.3075417518615723,
+      "mean_token_accuracy": 0.8964757239818573,
+      "num_tokens": 4769957.0,
+      "step": 2100
+    },
+    {
+      "entropy": 0.39732489734888077,
+      "epoch": 4.705366922234392,
+      "grad_norm": 1.3783643245697021,
+      "learning_rate": 7.548977734999099e-05,
+      "loss": 0.3155888748168945,
+      "mean_token_accuracy": 0.8941756916046143,
+      "num_tokens": 4882487.0,
+      "step": 2150
+    },
+    {
+      "entropy": 0.3972980257868767,
+      "epoch": 4.814895947426068,
+      "grad_norm": 0.9033240675926208,
+      "learning_rate": 7.330401974328283e-05,
+      "loss": 0.3149653244018555,
+      "mean_token_accuracy": 0.8941631782054901,
+      "num_tokens": 4993060.0,
+      "step": 2200
+    },
+    {
+      "entropy": 0.39401222407817843,
+      "epoch": 4.924424972617744,
+      "grad_norm": 1.1885665655136108,
+      "learning_rate": 7.109771589568177e-05,
+      "loss": 0.31299734115600586,
+      "mean_token_accuracy": 0.8955705845355988,
+      "num_tokens": 5111182.0,
+      "step": 2250
+    },
+    {
+      "epoch": 5.0,
+      "eval_entropy": 0.44022186218746123,
+      "eval_loss": 0.731863260269165,
+      "eval_mean_token_accuracy": 0.8086685883620429,
+      "eval_num_tokens": 5191250.0,
+      "eval_runtime": 44.8156,
+      "eval_samples_per_second": 22.358,
+      "eval_steps_per_second": 2.812,
+      "step": 2285
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 4570,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.1263896427218816e+17,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2742/README.md b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2742/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..7ac50e2706e5b4eae8f28e204601afb241c49e55
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2742/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-2B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-2B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2742/adapter_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2742/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..86d878f780798dd8f06b10d3bdb6eaee702b2da7
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2742/adapter_config.json
@@ -0,0 +1,48 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.01686122009693236,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "v_proj",
+    "o_proj",
+    "k_proj",
+    "up_proj",
+    "q_proj",
+    "down_proj",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2742/chat_template.jinja b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2742/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..0ef09f214eaa6d9bca297988afc1454b5827b2c7
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2742/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is true %}
+        {{- '<think>\n' }}
+    {%- else %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2742/tokenizer_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2742/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2742/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2742/trainer_state.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2742/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..730a3c2875779c54b6702914a23b59d130da1ad0
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-2742/trainer_state.json
@@ -0,0 +1,640 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 6.0,
+  "eval_steps": 500,
+  "global_step": 2742,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.074948468208313,
+      "epoch": 0.10952902519167579,
+      "grad_norm": 2.3511903285980225,
+      "learning_rate": 1.2698308676050528e-05,
+      "loss": 1.9153688049316406,
+      "mean_token_accuracy": 0.6077924159169197,
+      "num_tokens": 113087.0,
+      "step": 50
+    },
+    {
+      "entropy": 1.054056368470192,
+      "epoch": 0.21905805038335158,
+      "grad_norm": 2.3081915378570557,
+      "learning_rate": 2.5655766508755147e-05,
+      "loss": 0.9633240509033203,
+      "mean_token_accuracy": 0.741833764910698,
+      "num_tokens": 230392.0,
+      "step": 100
+    },
+    {
+      "entropy": 0.9184817910194397,
+      "epoch": 0.32858707557502737,
+      "grad_norm": 2.0965280532836914,
+      "learning_rate": 3.861322434145977e-05,
+      "loss": 0.8405924987792969,
+      "mean_token_accuracy": 0.7688284432888031,
+      "num_tokens": 341746.0,
+      "step": 150
+    },
+    {
+      "entropy": 0.8618861585855484,
+      "epoch": 0.43811610076670315,
+      "grad_norm": 2.2506542205810547,
+      "learning_rate": 5.157068217416438e-05,
+      "loss": 0.7841030883789063,
+      "mean_token_accuracy": 0.7796466761827469,
+      "num_tokens": 460483.0,
+      "step": 200
+    },
+    {
+      "entropy": 0.819869134426117,
+      "epoch": 0.547645125958379,
+      "grad_norm": 1.4058395624160767,
+      "learning_rate": 6.4528140006869e-05,
+      "loss": 0.7524736785888672,
+      "mean_token_accuracy": 0.7900535291433335,
+      "num_tokens": 570627.0,
+      "step": 250
+    },
+    {
+      "entropy": 0.8075837278366089,
+      "epoch": 0.6571741511500547,
+      "grad_norm": 1.5815014839172363,
+      "learning_rate": 7.748559783957362e-05,
+      "loss": 0.7388697814941406,
+      "mean_token_accuracy": 0.7914055424928665,
+      "num_tokens": 677217.0,
+      "step": 300
+    },
+    {
+      "entropy": 0.7868322026729584,
+      "epoch": 0.7667031763417306,
+      "grad_norm": 1.2377994060516357,
+      "learning_rate": 9.044305567227824e-05,
+      "loss": 0.7226168060302735,
+      "mean_token_accuracy": 0.7953901988267899,
+      "num_tokens": 790185.0,
+      "step": 350
+    },
+    {
+      "entropy": 0.7526325100660324,
+      "epoch": 0.8762322015334063,
+      "grad_norm": 1.203137755393982,
+      "learning_rate": 0.00010340051350498286,
+      "loss": 0.6991680908203125,
+      "mean_token_accuracy": 0.7992895567417144,
+      "num_tokens": 910080.0,
+      "step": 400
+    },
+    {
+      "entropy": 0.7626404595375061,
+      "epoch": 0.9857612267250822,
+      "grad_norm": 1.1625452041625977,
+      "learning_rate": 0.00011635797133768749,
+      "loss": 0.7055020904541016,
+      "mean_token_accuracy": 0.7997505795955658,
+      "num_tokens": 1023473.0,
+      "step": 450
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.7441067624659765,
+      "eval_loss": 0.7209385633468628,
+      "eval_mean_token_accuracy": 0.7895652093584575,
+      "eval_num_tokens": 1038250.0,
+      "eval_runtime": 45.5076,
+      "eval_samples_per_second": 22.018,
+      "eval_steps_per_second": 2.769,
+      "step": 457
+    },
+    {
+      "entropy": 0.720724637460227,
+      "epoch": 1.0941949616648412,
+      "grad_norm": 1.2537180185317993,
+      "learning_rate": 0.00011840069618947894,
+      "loss": 0.6586117553710937,
+      "mean_token_accuracy": 0.8081232917429221,
+      "num_tokens": 1128922.0,
+      "step": 500
+    },
+    {
+      "entropy": 0.7084310132265091,
+      "epoch": 1.203723986856517,
+      "grad_norm": 1.2347161769866943,
+      "learning_rate": 0.00011828501915154198,
+      "loss": 0.650611572265625,
+      "mean_token_accuracy": 0.808387525677681,
+      "num_tokens": 1245635.0,
+      "step": 550
+    },
+    {
+      "entropy": 0.706004958152771,
+      "epoch": 1.3132530120481927,
+      "grad_norm": 1.2140169143676758,
+      "learning_rate": 0.00011808319665683977,
+      "loss": 0.6442465209960937,
+      "mean_token_accuracy": 0.8122780376672745,
+      "num_tokens": 1354885.0,
+      "step": 600
+    },
+    {
+      "entropy": 0.6854291236400605,
+      "epoch": 1.4227820372398685,
+      "grad_norm": 1.1137559413909912,
+      "learning_rate": 0.00011779552303848118,
+      "loss": 0.6293138885498046,
+      "mean_token_accuracy": 0.8158071845769882,
+      "num_tokens": 1472232.0,
+      "step": 650
+    },
+    {
+      "entropy": 0.6885070586204529,
+      "epoch": 1.5323110624315444,
+      "grad_norm": 1.1074714660644531,
+      "learning_rate": 0.00011742241783280468,
+      "loss": 0.6259001541137695,
+      "mean_token_accuracy": 0.8164563828706741,
+      "num_tokens": 1587494.0,
+      "step": 700
+    },
+    {
+      "entropy": 0.6932114177942276,
+      "epoch": 1.6418400876232202,
+      "grad_norm": 1.1099495887756348,
+      "learning_rate": 0.00011696442516753647,
+      "loss": 0.6335408401489258,
+      "mean_token_accuracy": 0.8156683903932571,
+      "num_tokens": 1696927.0,
+      "step": 750
+    },
+    {
+      "entropy": 0.6787053138017655,
+      "epoch": 1.751369112814896,
+      "grad_norm": 1.2505418062210083,
+      "learning_rate": 0.00011642221296824764,
+      "loss": 0.6210909271240235,
+      "mean_token_accuracy": 0.8153067737817764,
+      "num_tokens": 1813415.0,
+      "step": 800
+    },
+    {
+      "entropy": 0.6693948286771775,
+      "epoch": 1.8608981380065717,
+      "grad_norm": 1.0963624715805054,
+      "learning_rate": 0.00011579657198426736,
+      "loss": 0.6105814361572266,
+      "mean_token_accuracy": 0.819609425663948,
+      "num_tokens": 1924911.0,
+      "step": 850
+    },
+    {
+      "entropy": 0.6634757363796234,
+      "epoch": 1.9704271631982475,
+      "grad_norm": 0.8766036033630371,
+      "learning_rate": 0.00011508841463547313,
+      "loss": 0.6100498580932617,
+      "mean_token_accuracy": 0.8197187691926956,
+      "num_tokens": 2044650.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.0,
+      "eval_entropy": 0.6342300275961558,
+      "eval_loss": 0.6517631411552429,
+      "eval_mean_token_accuracy": 0.807878240233376,
+      "eval_num_tokens": 2076500.0,
+      "eval_runtime": 44.6225,
+      "eval_samples_per_second": 22.455,
+      "eval_steps_per_second": 2.824,
+      "step": 914
+    },
+    {
+      "entropy": 0.6060671378867795,
+      "epoch": 2.078860898138007,
+      "grad_norm": 0.8028743267059326,
+      "learning_rate": 0.00011429877368163915,
+      "loss": 0.5440861511230469,
+      "mean_token_accuracy": 0.833196161973356,
+      "num_tokens": 2160498.0,
+      "step": 950
+    },
+    {
+      "entropy": 0.5839239662885666,
+      "epoch": 2.1883899233296824,
+      "grad_norm": 0.8585777878761292,
+      "learning_rate": 0.00011342880071628435,
+      "loss": 0.5182851409912109,
+      "mean_token_accuracy": 0.8406627786159515,
+      "num_tokens": 2269979.0,
+      "step": 1000
+    },
+    {
+      "entropy": 0.5949062475562096,
+      "epoch": 2.297918948521358,
+      "grad_norm": 0.9953155517578125,
+      "learning_rate": 0.00011247976448721578,
+      "loss": 0.5327045822143555,
+      "mean_token_accuracy": 0.8344444561004639,
+      "num_tokens": 2378362.0,
+      "step": 1050
+    },
+    {
+      "entropy": 0.583380132317543,
+      "epoch": 2.407447973713034,
+      "grad_norm": 0.9235143065452576,
+      "learning_rate": 0.00011145304904621704,
+      "loss": 0.520841178894043,
+      "mean_token_accuracy": 0.839934149980545,
+      "num_tokens": 2498494.0,
+      "step": 1100
+    },
+    {
+      "entropy": 0.5811308068037033,
+      "epoch": 2.5169769989047097,
+      "grad_norm": 1.1114075183868408,
+      "learning_rate": 0.00011035015173058038,
+      "loss": 0.5212027740478515,
+      "mean_token_accuracy": 0.838929146528244,
+      "num_tokens": 2611646.0,
+      "step": 1150
+    },
+    {
+      "entropy": 0.5903497302532196,
+      "epoch": 2.6265060240963853,
+      "grad_norm": 1.053895354270935,
+      "learning_rate": 0.0001091726809794255,
+      "loss": 0.5264106750488281,
+      "mean_token_accuracy": 0.8369076484441758,
+      "num_tokens": 2717744.0,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5949014449119567,
+      "epoch": 2.7360350492880614,
+      "grad_norm": 0.7563770413398743,
+      "learning_rate": 0.00010792235398799048,
+      "loss": 0.5322903442382813,
+      "mean_token_accuracy": 0.8368168956041336,
+      "num_tokens": 2827750.0,
+      "step": 1250
+    },
+    {
+      "entropy": 0.5745143675804139,
+      "epoch": 2.845564074479737,
+      "grad_norm": 0.8103823065757751,
+      "learning_rate": 0.00010660099420331526,
+      "loss": 0.5154821014404297,
+      "mean_token_accuracy": 0.8410378849506378,
+      "num_tokens": 2948958.0,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5665904894471169,
+      "epoch": 2.955093099671413,
+      "grad_norm": 0.8764206171035767,
+      "learning_rate": 0.00010521052866496986,
+      "loss": 0.5121800231933594,
+      "mean_token_accuracy": 0.8420848768949508,
+      "num_tokens": 3066917.0,
+      "step": 1350
+    },
+    {
+      "epoch": 3.0,
+      "eval_entropy": 0.5603872095308606,
+      "eval_loss": 0.6720245480537415,
+      "eval_mean_token_accuracy": 0.8079002830717299,
+      "eval_num_tokens": 3114750.0,
+      "eval_runtime": 43.9396,
+      "eval_samples_per_second": 22.804,
+      "eval_steps_per_second": 2.868,
+      "step": 1371
+    },
+    {
+      "entropy": 0.5181072778774031,
+      "epoch": 3.063526834611172,
+      "grad_norm": 0.889443039894104,
+      "learning_rate": 0.00010375298519470596,
+      "loss": 0.44869777679443357,
+      "mean_token_accuracy": 0.8574847211741438,
+      "num_tokens": 3180366.0,
+      "step": 1400
+    },
+    {
+      "entropy": 0.4902029836177826,
+      "epoch": 3.1730558598028478,
+      "grad_norm": 0.8999125957489014,
+      "learning_rate": 0.00010223048943913007,
+      "loss": 0.41646446228027345,
+      "mean_token_accuracy": 0.8655553787946701,
+      "num_tokens": 3293918.0,
+      "step": 1450
+    },
+    {
+      "entropy": 0.48207574665546415,
+      "epoch": 3.2825848849945234,
+      "grad_norm": 1.0042189359664917,
+      "learning_rate": 0.00010064526176971124,
+      "loss": 0.41205951690673825,
+      "mean_token_accuracy": 0.866786903142929,
+      "num_tokens": 3409058.0,
+      "step": 1500
+    },
+    {
+      "entropy": 0.4839185461401939,
+      "epoch": 3.3921139101861995,
+      "grad_norm": 1.1650396585464478,
+      "learning_rate": 9.89996140446443e-05,
+      "loss": 0.4102657699584961,
+      "mean_token_accuracy": 0.8668668657541275,
+      "num_tokens": 3525131.0,
+      "step": 1550
+    },
+    {
+      "entropy": 0.4932769918441772,
+      "epoch": 3.501642935377875,
+      "grad_norm": 1.0935224294662476,
+      "learning_rate": 9.729594623729099e-05,
+      "loss": 0.4225375366210937,
+      "mean_token_accuracy": 0.8634544950723648,
+      "num_tokens": 3641504.0,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5046219238638878,
+      "epoch": 3.6111719605695507,
+      "grad_norm": 0.9509316086769104,
+      "learning_rate": 9.553674293611626e-05,
+      "loss": 0.4319529342651367,
+      "mean_token_accuracy": 0.8610132175683975,
+      "num_tokens": 3752182.0,
+      "step": 1650
+    },
+    {
+      "entropy": 0.5090703725814819,
+      "epoch": 3.7207009857612268,
+      "grad_norm": 0.9050212502479553,
+      "learning_rate": 9.372456972122359e-05,
+      "loss": 0.4376397705078125,
+      "mean_token_accuracy": 0.8591927194595337,
+      "num_tokens": 3866473.0,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4903561845421791,
+      "epoch": 3.8302300109529024,
+      "grad_norm": 1.1624391078948975,
+      "learning_rate": 9.186206942277429e-05,
+      "loss": 0.4219248962402344,
+      "mean_token_accuracy": 0.8636374253034592,
+      "num_tokens": 3977393.0,
+      "step": 1750
+    },
+    {
+      "entropy": 0.5040662395954132,
+      "epoch": 3.9397590361445785,
+      "grad_norm": 0.798085629940033,
+      "learning_rate": 8.995195826674703e-05,
+      "loss": 0.4333121871948242,
+      "mean_token_accuracy": 0.86187963783741,
+      "num_tokens": 4088936.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5136857172326436,
+      "eval_loss": 0.6814945340156555,
+      "eval_mean_token_accuracy": 0.8091600262929523,
+      "eval_num_tokens": 4153000.0,
+      "eval_runtime": 44.5059,
+      "eval_samples_per_second": 22.514,
+      "eval_steps_per_second": 2.831,
+      "step": 1828
+    },
+    {
+      "entropy": 0.4410170723091472,
+      "epoch": 4.048192771084337,
+      "grad_norm": 0.9169163703918457,
+      "learning_rate": 8.799702191365879e-05,
+      "loss": 0.3653216552734375,
+      "mean_token_accuracy": 0.8799899684058295,
+      "num_tokens": 4202329.0,
+      "step": 1850
+    },
+    {
+      "entropy": 0.38991558194160464,
+      "epoch": 4.157721796276014,
+      "grad_norm": 1.2269299030303955,
+      "learning_rate": 8.600011139602396e-05,
+      "loss": 0.3035482597351074,
+      "mean_token_accuracy": 0.8981181675195694,
+      "num_tokens": 4320321.0,
+      "step": 1900
+    },
+    {
+      "entropy": 0.38429499566555025,
+      "epoch": 4.267250821467689,
+      "grad_norm": 1.1254630088806152,
+      "learning_rate": 8.39641389604766e-05,
+      "loss": 0.3037366676330566,
+      "mean_token_accuracy": 0.8979371869564057,
+      "num_tokens": 4435139.0,
+      "step": 1950
+    },
+    {
+      "entropy": 0.39086049795150757,
+      "epoch": 4.376779846659365,
+      "grad_norm": 1.3771028518676758,
+      "learning_rate": 8.18920738206195e-05,
+      "loss": 0.3100498390197754,
+      "mean_token_accuracy": 0.8953604429960251,
+      "num_tokens": 4547272.0,
+      "step": 2000
+    },
+    {
+      "entropy": 0.3967577290534973,
+      "epoch": 4.48630887185104,
+      "grad_norm": 1.0673140287399292,
+      "learning_rate": 7.978693782679394e-05,
+      "loss": 0.31265506744384763,
+      "mean_token_accuracy": 0.8946040958166123,
+      "num_tokens": 4659036.0,
+      "step": 2050
+    },
+    {
+      "entropy": 0.3927252873778343,
+      "epoch": 4.595837897042716,
+      "grad_norm": 1.1611689329147339,
+      "learning_rate": 7.765180105908515e-05,
+      "loss": 0.3075417518615723,
+      "mean_token_accuracy": 0.8964757239818573,
+      "num_tokens": 4769957.0,
+      "step": 2100
+    },
+    {
+      "entropy": 0.39732489734888077,
+      "epoch": 4.705366922234392,
+      "grad_norm": 1.3783643245697021,
+      "learning_rate": 7.548977734999099e-05,
+      "loss": 0.3155888748168945,
+      "mean_token_accuracy": 0.8941756916046143,
+      "num_tokens": 4882487.0,
+      "step": 2150
+    },
+    {
+      "entropy": 0.3972980257868767,
+      "epoch": 4.814895947426068,
+      "grad_norm": 0.9033240675926208,
+      "learning_rate": 7.330401974328283e-05,
+      "loss": 0.3149653244018555,
+      "mean_token_accuracy": 0.8941631782054901,
+      "num_tokens": 4993060.0,
+      "step": 2200
+    },
+    {
+      "entropy": 0.39401222407817843,
+      "epoch": 4.924424972617744,
+      "grad_norm": 1.1885665655136108,
+      "learning_rate": 7.109771589568177e-05,
+      "loss": 0.31299734115600586,
+      "mean_token_accuracy": 0.8955705845355988,
+      "num_tokens": 5111182.0,
+      "step": 2250
+    },
+    {
+      "epoch": 5.0,
+      "eval_entropy": 0.44022186218746123,
+      "eval_loss": 0.731863260269165,
+      "eval_mean_token_accuracy": 0.8086685883620429,
+      "eval_num_tokens": 5191250.0,
+      "eval_runtime": 44.8156,
+      "eval_samples_per_second": 22.358,
+      "eval_steps_per_second": 2.812,
+      "step": 2285
+    },
+    {
+      "entropy": 0.3665939736546892,
+      "epoch": 5.032858707557502,
+      "grad_norm": 1.2354992628097534,
+      "learning_rate": 6.887408342805626e-05,
+      "loss": 0.27955495834350585,
+      "mean_token_accuracy": 0.9059046464737015,
+      "num_tokens": 5225197.0,
+      "step": 2300
+    },
+    {
+      "entropy": 0.2878261508047581,
+      "epoch": 5.142387732749179,
+      "grad_norm": 1.3155502080917358,
+      "learning_rate": 6.663636523292057e-05,
+      "loss": 0.1942205810546875,
+      "mean_token_accuracy": 0.9343426609039307,
+      "num_tokens": 5336728.0,
+      "step": 2350
+    },
+    {
+      "entropy": 0.2881846517324448,
+      "epoch": 5.2519167579408546,
+      "grad_norm": 1.157877802848816,
+      "learning_rate": 6.438782474507771e-05,
+      "loss": 0.19725875854492186,
+      "mean_token_accuracy": 0.9327778881788253,
+      "num_tokens": 5446986.0,
+      "step": 2400
+    },
+    {
+      "entropy": 0.2804257667064667,
+      "epoch": 5.36144578313253,
+      "grad_norm": 1.3322943449020386,
+      "learning_rate": 6.213174118230412e-05,
+      "loss": 0.19409126281738281,
+      "mean_token_accuracy": 0.9335259759426117,
+      "num_tokens": 5563479.0,
+      "step": 2450
+    },
+    {
+      "entropy": 0.28692082852125167,
+      "epoch": 5.470974808324206,
+      "grad_norm": 1.2746684551239014,
+      "learning_rate": 5.987140476301658e-05,
+      "loss": 0.2002597999572754,
+      "mean_token_accuracy": 0.9313689643144607,
+      "num_tokens": 5677338.0,
+      "step": 2500
+    },
+    {
+      "entropy": 0.2895388266444206,
+      "epoch": 5.580503833515881,
+      "grad_norm": 1.149001121520996,
+      "learning_rate": 5.761011190789638e-05,
+      "loss": 0.2018396759033203,
+      "mean_token_accuracy": 0.9323018848896026,
+      "num_tokens": 5787988.0,
+      "step": 2550
+    },
+    {
+      "entropy": 0.29092250213027,
+      "epoch": 5.690032858707557,
+      "grad_norm": 1.3538148403167725,
+      "learning_rate": 5.5351160432467955e-05,
+      "loss": 0.20052331924438477,
+      "mean_token_accuracy": 0.9314265781641007,
+      "num_tokens": 5902559.0,
+      "step": 2600
+    },
+    {
+      "entropy": 0.28983849346637724,
+      "epoch": 5.7995618838992335,
+      "grad_norm": 1.2387913465499878,
+      "learning_rate": 5.309784473764352e-05,
+      "loss": 0.20004833221435547,
+      "mean_token_accuracy": 0.9308328247070312,
+      "num_tokens": 6014039.0,
+      "step": 2650
+    },
+    {
+      "entropy": 0.27550786405801775,
+      "epoch": 5.909090909090909,
+      "grad_norm": 1.1062148809432983,
+      "learning_rate": 5.085345100524733e-05,
+      "loss": 0.19400547027587892,
+      "mean_token_accuracy": 0.9349210584163665,
+      "num_tokens": 6134464.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.0,
+      "eval_entropy": 0.36818124211969827,
+      "eval_loss": 0.8581439256668091,
+      "eval_mean_token_accuracy": 0.8029990144192226,
+      "eval_num_tokens": 6229500.0,
+      "eval_runtime": 44.8208,
+      "eval_samples_per_second": 22.356,
+      "eval_steps_per_second": 2.811,
+      "step": 2742
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 4570,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.3508240629976256e+17,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3199/README.md b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3199/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..7ac50e2706e5b4eae8f28e204601afb241c49e55
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3199/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-2B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-2B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3199/adapter_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3199/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..86d878f780798dd8f06b10d3bdb6eaee702b2da7
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3199/adapter_config.json
@@ -0,0 +1,48 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.01686122009693236,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "v_proj",
+    "o_proj",
+    "k_proj",
+    "up_proj",
+    "q_proj",
+    "down_proj",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3199/chat_template.jinja b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3199/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..0ef09f214eaa6d9bca297988afc1454b5827b2c7
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3199/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is true %}
+        {{- '<think>\n' }}
+    {%- else %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3199/tokenizer_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3199/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3199/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3199/trainer_state.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3199/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..7084de3e22bb6eeb8ff4d3aa9edc3503e46f5c3c
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3199/trainer_state.json
@@ -0,0 +1,741 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 7.0,
+  "eval_steps": 500,
+  "global_step": 3199,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.074948468208313,
+      "epoch": 0.10952902519167579,
+      "grad_norm": 2.3511903285980225,
+      "learning_rate": 1.2698308676050528e-05,
+      "loss": 1.9153688049316406,
+      "mean_token_accuracy": 0.6077924159169197,
+      "num_tokens": 113087.0,
+      "step": 50
+    },
+    {
+      "entropy": 1.054056368470192,
+      "epoch": 0.21905805038335158,
+      "grad_norm": 2.3081915378570557,
+      "learning_rate": 2.5655766508755147e-05,
+      "loss": 0.9633240509033203,
+      "mean_token_accuracy": 0.741833764910698,
+      "num_tokens": 230392.0,
+      "step": 100
+    },
+    {
+      "entropy": 0.9184817910194397,
+      "epoch": 0.32858707557502737,
+      "grad_norm": 2.0965280532836914,
+      "learning_rate": 3.861322434145977e-05,
+      "loss": 0.8405924987792969,
+      "mean_token_accuracy": 0.7688284432888031,
+      "num_tokens": 341746.0,
+      "step": 150
+    },
+    {
+      "entropy": 0.8618861585855484,
+      "epoch": 0.43811610076670315,
+      "grad_norm": 2.2506542205810547,
+      "learning_rate": 5.157068217416438e-05,
+      "loss": 0.7841030883789063,
+      "mean_token_accuracy": 0.7796466761827469,
+      "num_tokens": 460483.0,
+      "step": 200
+    },
+    {
+      "entropy": 0.819869134426117,
+      "epoch": 0.547645125958379,
+      "grad_norm": 1.4058395624160767,
+      "learning_rate": 6.4528140006869e-05,
+      "loss": 0.7524736785888672,
+      "mean_token_accuracy": 0.7900535291433335,
+      "num_tokens": 570627.0,
+      "step": 250
+    },
+    {
+      "entropy": 0.8075837278366089,
+      "epoch": 0.6571741511500547,
+      "grad_norm": 1.5815014839172363,
+      "learning_rate": 7.748559783957362e-05,
+      "loss": 0.7388697814941406,
+      "mean_token_accuracy": 0.7914055424928665,
+      "num_tokens": 677217.0,
+      "step": 300
+    },
+    {
+      "entropy": 0.7868322026729584,
+      "epoch": 0.7667031763417306,
+      "grad_norm": 1.2377994060516357,
+      "learning_rate": 9.044305567227824e-05,
+      "loss": 0.7226168060302735,
+      "mean_token_accuracy": 0.7953901988267899,
+      "num_tokens": 790185.0,
+      "step": 350
+    },
+    {
+      "entropy": 0.7526325100660324,
+      "epoch": 0.8762322015334063,
+      "grad_norm": 1.203137755393982,
+      "learning_rate": 0.00010340051350498286,
+      "loss": 0.6991680908203125,
+      "mean_token_accuracy": 0.7992895567417144,
+      "num_tokens": 910080.0,
+      "step": 400
+    },
+    {
+      "entropy": 0.7626404595375061,
+      "epoch": 0.9857612267250822,
+      "grad_norm": 1.1625452041625977,
+      "learning_rate": 0.00011635797133768749,
+      "loss": 0.7055020904541016,
+      "mean_token_accuracy": 0.7997505795955658,
+      "num_tokens": 1023473.0,
+      "step": 450
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.7441067624659765,
+      "eval_loss": 0.7209385633468628,
+      "eval_mean_token_accuracy": 0.7895652093584575,
+      "eval_num_tokens": 1038250.0,
+      "eval_runtime": 45.5076,
+      "eval_samples_per_second": 22.018,
+      "eval_steps_per_second": 2.769,
+      "step": 457
+    },
+    {
+      "entropy": 0.720724637460227,
+      "epoch": 1.0941949616648412,
+      "grad_norm": 1.2537180185317993,
+      "learning_rate": 0.00011840069618947894,
+      "loss": 0.6586117553710937,
+      "mean_token_accuracy": 0.8081232917429221,
+      "num_tokens": 1128922.0,
+      "step": 500
+    },
+    {
+      "entropy": 0.7084310132265091,
+      "epoch": 1.203723986856517,
+      "grad_norm": 1.2347161769866943,
+      "learning_rate": 0.00011828501915154198,
+      "loss": 0.650611572265625,
+      "mean_token_accuracy": 0.808387525677681,
+      "num_tokens": 1245635.0,
+      "step": 550
+    },
+    {
+      "entropy": 0.706004958152771,
+      "epoch": 1.3132530120481927,
+      "grad_norm": 1.2140169143676758,
+      "learning_rate": 0.00011808319665683977,
+      "loss": 0.6442465209960937,
+      "mean_token_accuracy": 0.8122780376672745,
+      "num_tokens": 1354885.0,
+      "step": 600
+    },
+    {
+      "entropy": 0.6854291236400605,
+      "epoch": 1.4227820372398685,
+      "grad_norm": 1.1137559413909912,
+      "learning_rate": 0.00011779552303848118,
+      "loss": 0.6293138885498046,
+      "mean_token_accuracy": 0.8158071845769882,
+      "num_tokens": 1472232.0,
+      "step": 650
+    },
+    {
+      "entropy": 0.6885070586204529,
+      "epoch": 1.5323110624315444,
+      "grad_norm": 1.1074714660644531,
+      "learning_rate": 0.00011742241783280468,
+      "loss": 0.6259001541137695,
+      "mean_token_accuracy": 0.8164563828706741,
+      "num_tokens": 1587494.0,
+      "step": 700
+    },
+    {
+      "entropy": 0.6932114177942276,
+      "epoch": 1.6418400876232202,
+      "grad_norm": 1.1099495887756348,
+      "learning_rate": 0.00011696442516753647,
+      "loss": 0.6335408401489258,
+      "mean_token_accuracy": 0.8156683903932571,
+      "num_tokens": 1696927.0,
+      "step": 750
+    },
+    {
+      "entropy": 0.6787053138017655,
+      "epoch": 1.751369112814896,
+      "grad_norm": 1.2505418062210083,
+      "learning_rate": 0.00011642221296824764,
+      "loss": 0.6210909271240235,
+      "mean_token_accuracy": 0.8153067737817764,
+      "num_tokens": 1813415.0,
+      "step": 800
+    },
+    {
+      "entropy": 0.6693948286771775,
+      "epoch": 1.8608981380065717,
+      "grad_norm": 1.0963624715805054,
+      "learning_rate": 0.00011579657198426736,
+      "loss": 0.6105814361572266,
+      "mean_token_accuracy": 0.819609425663948,
+      "num_tokens": 1924911.0,
+      "step": 850
+    },
+    {
+      "entropy": 0.6634757363796234,
+      "epoch": 1.9704271631982475,
+      "grad_norm": 0.8766036033630371,
+      "learning_rate": 0.00011508841463547313,
+      "loss": 0.6100498580932617,
+      "mean_token_accuracy": 0.8197187691926956,
+      "num_tokens": 2044650.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.0,
+      "eval_entropy": 0.6342300275961558,
+      "eval_loss": 0.6517631411552429,
+      "eval_mean_token_accuracy": 0.807878240233376,
+      "eval_num_tokens": 2076500.0,
+      "eval_runtime": 44.6225,
+      "eval_samples_per_second": 22.455,
+      "eval_steps_per_second": 2.824,
+      "step": 914
+    },
+    {
+      "entropy": 0.6060671378867795,
+      "epoch": 2.078860898138007,
+      "grad_norm": 0.8028743267059326,
+      "learning_rate": 0.00011429877368163915,
+      "loss": 0.5440861511230469,
+      "mean_token_accuracy": 0.833196161973356,
+      "num_tokens": 2160498.0,
+      "step": 950
+    },
+    {
+      "entropy": 0.5839239662885666,
+      "epoch": 2.1883899233296824,
+      "grad_norm": 0.8585777878761292,
+      "learning_rate": 0.00011342880071628435,
+      "loss": 0.5182851409912109,
+      "mean_token_accuracy": 0.8406627786159515,
+      "num_tokens": 2269979.0,
+      "step": 1000
+    },
+    {
+      "entropy": 0.5949062475562096,
+      "epoch": 2.297918948521358,
+      "grad_norm": 0.9953155517578125,
+      "learning_rate": 0.00011247976448721578,
+      "loss": 0.5327045822143555,
+      "mean_token_accuracy": 0.8344444561004639,
+      "num_tokens": 2378362.0,
+      "step": 1050
+    },
+    {
+      "entropy": 0.583380132317543,
+      "epoch": 2.407447973713034,
+      "grad_norm": 0.9235143065452576,
+      "learning_rate": 0.00011145304904621704,
+      "loss": 0.520841178894043,
+      "mean_token_accuracy": 0.839934149980545,
+      "num_tokens": 2498494.0,
+      "step": 1100
+    },
+    {
+      "entropy": 0.5811308068037033,
+      "epoch": 2.5169769989047097,
+      "grad_norm": 1.1114075183868408,
+      "learning_rate": 0.00011035015173058038,
+      "loss": 0.5212027740478515,
+      "mean_token_accuracy": 0.838929146528244,
+      "num_tokens": 2611646.0,
+      "step": 1150
+    },
+    {
+      "entropy": 0.5903497302532196,
+      "epoch": 2.6265060240963853,
+      "grad_norm": 1.053895354270935,
+      "learning_rate": 0.0001091726809794255,
+      "loss": 0.5264106750488281,
+      "mean_token_accuracy": 0.8369076484441758,
+      "num_tokens": 2717744.0,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5949014449119567,
+      "epoch": 2.7360350492880614,
+      "grad_norm": 0.7563770413398743,
+      "learning_rate": 0.00010792235398799048,
+      "loss": 0.5322903442382813,
+      "mean_token_accuracy": 0.8368168956041336,
+      "num_tokens": 2827750.0,
+      "step": 1250
+    },
+    {
+      "entropy": 0.5745143675804139,
+      "epoch": 2.845564074479737,
+      "grad_norm": 0.8103823065757751,
+      "learning_rate": 0.00010660099420331526,
+      "loss": 0.5154821014404297,
+      "mean_token_accuracy": 0.8410378849506378,
+      "num_tokens": 2948958.0,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5665904894471169,
+      "epoch": 2.955093099671413,
+      "grad_norm": 0.8764206171035767,
+      "learning_rate": 0.00010521052866496986,
+      "loss": 0.5121800231933594,
+      "mean_token_accuracy": 0.8420848768949508,
+      "num_tokens": 3066917.0,
+      "step": 1350
+    },
+    {
+      "epoch": 3.0,
+      "eval_entropy": 0.5603872095308606,
+      "eval_loss": 0.6720245480537415,
+      "eval_mean_token_accuracy": 0.8079002830717299,
+      "eval_num_tokens": 3114750.0,
+      "eval_runtime": 43.9396,
+      "eval_samples_per_second": 22.804,
+      "eval_steps_per_second": 2.868,
+      "step": 1371
+    },
+    {
+      "entropy": 0.5181072778774031,
+      "epoch": 3.063526834611172,
+      "grad_norm": 0.889443039894104,
+      "learning_rate": 0.00010375298519470596,
+      "loss": 0.44869777679443357,
+      "mean_token_accuracy": 0.8574847211741438,
+      "num_tokens": 3180366.0,
+      "step": 1400
+    },
+    {
+      "entropy": 0.4902029836177826,
+      "epoch": 3.1730558598028478,
+      "grad_norm": 0.8999125957489014,
+      "learning_rate": 0.00010223048943913007,
+      "loss": 0.41646446228027345,
+      "mean_token_accuracy": 0.8655553787946701,
+      "num_tokens": 3293918.0,
+      "step": 1450
+    },
+    {
+      "entropy": 0.48207574665546415,
+      "epoch": 3.2825848849945234,
+      "grad_norm": 1.0042189359664917,
+      "learning_rate": 0.00010064526176971124,
+      "loss": 0.41205951690673825,
+      "mean_token_accuracy": 0.866786903142929,
+      "num_tokens": 3409058.0,
+      "step": 1500
+    },
+    {
+      "entropy": 0.4839185461401939,
+      "epoch": 3.3921139101861995,
+      "grad_norm": 1.1650396585464478,
+      "learning_rate": 9.89996140446443e-05,
+      "loss": 0.4102657699584961,
+      "mean_token_accuracy": 0.8668668657541275,
+      "num_tokens": 3525131.0,
+      "step": 1550
+    },
+    {
+      "entropy": 0.4932769918441772,
+      "epoch": 3.501642935377875,
+      "grad_norm": 1.0935224294662476,
+      "learning_rate": 9.729594623729099e-05,
+      "loss": 0.4225375366210937,
+      "mean_token_accuracy": 0.8634544950723648,
+      "num_tokens": 3641504.0,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5046219238638878,
+      "epoch": 3.6111719605695507,
+      "grad_norm": 0.9509316086769104,
+      "learning_rate": 9.553674293611626e-05,
+      "loss": 0.4319529342651367,
+      "mean_token_accuracy": 0.8610132175683975,
+      "num_tokens": 3752182.0,
+      "step": 1650
+    },
+    {
+      "entropy": 0.5090703725814819,
+      "epoch": 3.7207009857612268,
+      "grad_norm": 0.9050212502479553,
+      "learning_rate": 9.372456972122359e-05,
+      "loss": 0.4376397705078125,
+      "mean_token_accuracy": 0.8591927194595337,
+      "num_tokens": 3866473.0,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4903561845421791,
+      "epoch": 3.8302300109529024,
+      "grad_norm": 1.1624391078948975,
+      "learning_rate": 9.186206942277429e-05,
+      "loss": 0.4219248962402344,
+      "mean_token_accuracy": 0.8636374253034592,
+      "num_tokens": 3977393.0,
+      "step": 1750
+    },
+    {
+      "entropy": 0.5040662395954132,
+      "epoch": 3.9397590361445785,
+      "grad_norm": 0.798085629940033,
+      "learning_rate": 8.995195826674703e-05,
+      "loss": 0.4333121871948242,
+      "mean_token_accuracy": 0.86187963783741,
+      "num_tokens": 4088936.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5136857172326436,
+      "eval_loss": 0.6814945340156555,
+      "eval_mean_token_accuracy": 0.8091600262929523,
+      "eval_num_tokens": 4153000.0,
+      "eval_runtime": 44.5059,
+      "eval_samples_per_second": 22.514,
+      "eval_steps_per_second": 2.831,
+      "step": 1828
+    },
+    {
+      "entropy": 0.4410170723091472,
+      "epoch": 4.048192771084337,
+      "grad_norm": 0.9169163703918457,
+      "learning_rate": 8.799702191365879e-05,
+      "loss": 0.3653216552734375,
+      "mean_token_accuracy": 0.8799899684058295,
+      "num_tokens": 4202329.0,
+      "step": 1850
+    },
+    {
+      "entropy": 0.38991558194160464,
+      "epoch": 4.157721796276014,
+      "grad_norm": 1.2269299030303955,
+      "learning_rate": 8.600011139602396e-05,
+      "loss": 0.3035482597351074,
+      "mean_token_accuracy": 0.8981181675195694,
+      "num_tokens": 4320321.0,
+      "step": 1900
+    },
+    {
+      "entropy": 0.38429499566555025,
+      "epoch": 4.267250821467689,
+      "grad_norm": 1.1254630088806152,
+      "learning_rate": 8.39641389604766e-05,
+      "loss": 0.3037366676330566,
+      "mean_token_accuracy": 0.8979371869564057,
+      "num_tokens": 4435139.0,
+      "step": 1950
+    },
+    {
+      "entropy": 0.39086049795150757,
+      "epoch": 4.376779846659365,
+      "grad_norm": 1.3771028518676758,
+      "learning_rate": 8.18920738206195e-05,
+      "loss": 0.3100498390197754,
+      "mean_token_accuracy": 0.8953604429960251,
+      "num_tokens": 4547272.0,
+      "step": 2000
+    },
+    {
+      "entropy": 0.3967577290534973,
+      "epoch": 4.48630887185104,
+      "grad_norm": 1.0673140287399292,
+      "learning_rate": 7.978693782679394e-05,
+      "loss": 0.31265506744384763,
+      "mean_token_accuracy": 0.8946040958166123,
+      "num_tokens": 4659036.0,
+      "step": 2050
+    },
+    {
+      "entropy": 0.3927252873778343,
+      "epoch": 4.595837897042716,
+      "grad_norm": 1.1611689329147339,
+      "learning_rate": 7.765180105908515e-05,
+      "loss": 0.3075417518615723,
+      "mean_token_accuracy": 0.8964757239818573,
+      "num_tokens": 4769957.0,
+      "step": 2100
+    },
+    {
+      "entropy": 0.39732489734888077,
+      "epoch": 4.705366922234392,
+      "grad_norm": 1.3783643245697021,
+      "learning_rate": 7.548977734999099e-05,
+      "loss": 0.3155888748168945,
+      "mean_token_accuracy": 0.8941756916046143,
+      "num_tokens": 4882487.0,
+      "step": 2150
+    },
+    {
+      "entropy": 0.3972980257868767,
+      "epoch": 4.814895947426068,
+      "grad_norm": 0.9033240675926208,
+      "learning_rate": 7.330401974328283e-05,
+      "loss": 0.3149653244018555,
+      "mean_token_accuracy": 0.8941631782054901,
+      "num_tokens": 4993060.0,
+      "step": 2200
+    },
+    {
+      "entropy": 0.39401222407817843,
+      "epoch": 4.924424972617744,
+      "grad_norm": 1.1885665655136108,
+      "learning_rate": 7.109771589568177e-05,
+      "loss": 0.31299734115600586,
+      "mean_token_accuracy": 0.8955705845355988,
+      "num_tokens": 5111182.0,
+      "step": 2250
+    },
+    {
+      "epoch": 5.0,
+      "eval_entropy": 0.44022186218746123,
+      "eval_loss": 0.731863260269165,
+      "eval_mean_token_accuracy": 0.8086685883620429,
+      "eval_num_tokens": 5191250.0,
+      "eval_runtime": 44.8156,
+      "eval_samples_per_second": 22.358,
+      "eval_steps_per_second": 2.812,
+      "step": 2285
+    },
+    {
+      "entropy": 0.3665939736546892,
+      "epoch": 5.032858707557502,
+      "grad_norm": 1.2354992628097534,
+      "learning_rate": 6.887408342805626e-05,
+      "loss": 0.27955495834350585,
+      "mean_token_accuracy": 0.9059046464737015,
+      "num_tokens": 5225197.0,
+      "step": 2300
+    },
+    {
+      "entropy": 0.2878261508047581,
+      "epoch": 5.142387732749179,
+      "grad_norm": 1.3155502080917358,
+      "learning_rate": 6.663636523292057e-05,
+      "loss": 0.1942205810546875,
+      "mean_token_accuracy": 0.9343426609039307,
+      "num_tokens": 5336728.0,
+      "step": 2350
+    },
+    {
+      "entropy": 0.2881846517324448,
+      "epoch": 5.2519167579408546,
+      "grad_norm": 1.157877802848816,
+      "learning_rate": 6.438782474507771e-05,
+      "loss": 0.19725875854492186,
+      "mean_token_accuracy": 0.9327778881788253,
+      "num_tokens": 5446986.0,
+      "step": 2400
+    },
+    {
+      "entropy": 0.2804257667064667,
+      "epoch": 5.36144578313253,
+      "grad_norm": 1.3322943449020386,
+      "learning_rate": 6.213174118230412e-05,
+      "loss": 0.19409126281738281,
+      "mean_token_accuracy": 0.9335259759426117,
+      "num_tokens": 5563479.0,
+      "step": 2450
+    },
+    {
+      "entropy": 0.28692082852125167,
+      "epoch": 5.470974808324206,
+      "grad_norm": 1.2746684551239014,
+      "learning_rate": 5.987140476301658e-05,
+      "loss": 0.2002597999572754,
+      "mean_token_accuracy": 0.9313689643144607,
+      "num_tokens": 5677338.0,
+      "step": 2500
+    },
+    {
+      "entropy": 0.2895388266444206,
+      "epoch": 5.580503833515881,
+      "grad_norm": 1.149001121520996,
+      "learning_rate": 5.761011190789638e-05,
+      "loss": 0.2018396759033203,
+      "mean_token_accuracy": 0.9323018848896026,
+      "num_tokens": 5787988.0,
+      "step": 2550
+    },
+    {
+      "entropy": 0.29092250213027,
+      "epoch": 5.690032858707557,
+      "grad_norm": 1.3538148403167725,
+      "learning_rate": 5.5351160432467955e-05,
+      "loss": 0.20052331924438477,
+      "mean_token_accuracy": 0.9314265781641007,
+      "num_tokens": 5902559.0,
+      "step": 2600
+    },
+    {
+      "entropy": 0.28983849346637724,
+      "epoch": 5.7995618838992335,
+      "grad_norm": 1.2387913465499878,
+      "learning_rate": 5.309784473764352e-05,
+      "loss": 0.20004833221435547,
+      "mean_token_accuracy": 0.9308328247070312,
+      "num_tokens": 6014039.0,
+      "step": 2650
+    },
+    {
+      "entropy": 0.27550786405801775,
+      "epoch": 5.909090909090909,
+      "grad_norm": 1.1062148809432983,
+      "learning_rate": 5.085345100524733e-05,
+      "loss": 0.19400547027587892,
+      "mean_token_accuracy": 0.9349210584163665,
+      "num_tokens": 6134464.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.0,
+      "eval_entropy": 0.36818124211969827,
+      "eval_loss": 0.8581439256668091,
+      "eval_mean_token_accuracy": 0.8029990144192226,
+      "eval_num_tokens": 6229500.0,
+      "eval_runtime": 44.8208,
+      "eval_samples_per_second": 22.356,
+      "eval_steps_per_second": 2.811,
+      "step": 2742
+    },
+    {
+      "entropy": 0.2698795477850269,
+      "epoch": 6.017524644030668,
+      "grad_norm": 0.9266197085380554,
+      "learning_rate": 4.862125240552664e-05,
+      "loss": 0.18622320175170898,
+      "mean_token_accuracy": 0.9371824848531473,
+      "num_tokens": 6249905.0,
+      "step": 2750
+    },
+    {
+      "entropy": 0.19044648699462413,
+      "epoch": 6.127053669222344,
+      "grad_norm": 1.0554289817810059,
+      "learning_rate": 4.640450432363826e-05,
+      "loss": 0.11102401733398437,
+      "mean_token_accuracy": 0.9636328971385956,
+      "num_tokens": 6367665.0,
+      "step": 2800
+    },
+    {
+      "entropy": 0.1951981072127819,
+      "epoch": 6.23658269441402,
+      "grad_norm": 1.1141842603683472,
+      "learning_rate": 4.420643961207262e-05,
+      "loss": 0.11578875541687011,
+      "mean_token_accuracy": 0.9620302796363831,
+      "num_tokens": 6481769.0,
+      "step": 2850
+    },
+    {
+      "entropy": 0.20094555169343947,
+      "epoch": 6.3461117196056955,
+      "grad_norm": 0.9116718769073486,
+      "learning_rate": 4.2030263875938836e-05,
+      "loss": 0.11906105995178223,
+      "mean_token_accuracy": 0.9608682131767273,
+      "num_tokens": 6595712.0,
+      "step": 2900
+    },
+    {
+      "entropy": 0.19917954877018929,
+      "epoch": 6.455640744797371,
+      "grad_norm": 1.0561001300811768,
+      "learning_rate": 3.98791507979867e-05,
+      "loss": 0.11807474136352539,
+      "mean_token_accuracy": 0.9615647631883621,
+      "num_tokens": 6708841.0,
+      "step": 2950
+    },
+    {
+      "entropy": 0.20109083890914917,
+      "epoch": 6.565169769989047,
+      "grad_norm": 0.9614781141281128,
+      "learning_rate": 3.7756237510183554e-05,
+      "loss": 0.12305583000183105,
+      "mean_token_accuracy": 0.9595666688680649,
+      "num_tokens": 6818025.0,
+      "step": 3000
+    },
+    {
+      "entropy": 0.20449081182479859,
+      "epoch": 6.674698795180722,
+      "grad_norm": 1.1171538829803467,
+      "learning_rate": 3.566462001859587e-05,
+      "loss": 0.12225373268127442,
+      "mean_token_accuracy": 0.9595645654201508,
+      "num_tokens": 6929746.0,
+      "step": 3050
+    },
+    {
+      "entropy": 0.19336393870413304,
+      "epoch": 6.784227820372399,
+      "grad_norm": 0.9689428806304932,
+      "learning_rate": 3.3607348688248006e-05,
+      "loss": 0.11655777931213379,
+      "mean_token_accuracy": 0.9615726363658905,
+      "num_tokens": 7046109.0,
+      "step": 3100
+    },
+    {
+      "entropy": 0.20375137731432916,
+      "epoch": 6.8937568455640745,
+      "grad_norm": 1.1648379564285278,
+      "learning_rate": 3.158742379454244e-05,
+      "loss": 0.12166361808776856,
+      "mean_token_accuracy": 0.9593629771471024,
+      "num_tokens": 7156155.0,
+      "step": 3150
+    },
+    {
+      "epoch": 7.0,
+      "eval_entropy": 0.31269029941823745,
+      "eval_loss": 1.0066559314727783,
+      "eval_mean_token_accuracy": 0.802609798454103,
+      "eval_num_tokens": 7267750.0,
+      "eval_runtime": 44.8168,
+      "eval_samples_per_second": 22.358,
+      "eval_steps_per_second": 2.811,
+      "step": 3199
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 4570,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.5761649650331648e+17,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3656/README.md b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3656/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..7ac50e2706e5b4eae8f28e204601afb241c49e55
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3656/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-2B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-2B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3656/adapter_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3656/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..86d878f780798dd8f06b10d3bdb6eaee702b2da7
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3656/adapter_config.json
@@ -0,0 +1,48 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.01686122009693236,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "v_proj",
+    "o_proj",
+    "k_proj",
+    "up_proj",
+    "q_proj",
+    "down_proj",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3656/chat_template.jinja b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3656/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..0ef09f214eaa6d9bca297988afc1454b5827b2c7
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3656/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is true %}
+        {{- '<think>\n' }}
+    {%- else %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3656/tokenizer_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3656/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3656/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3656/trainer_state.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3656/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..43c101fd36eb8b72ea1a9012449622dbe579fc4d
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-3656/trainer_state.json
@@ -0,0 +1,852 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.0,
+  "eval_steps": 500,
+  "global_step": 3656,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.074948468208313,
+      "epoch": 0.10952902519167579,
+      "grad_norm": 2.3511903285980225,
+      "learning_rate": 1.2698308676050528e-05,
+      "loss": 1.9153688049316406,
+      "mean_token_accuracy": 0.6077924159169197,
+      "num_tokens": 113087.0,
+      "step": 50
+    },
+    {
+      "entropy": 1.054056368470192,
+      "epoch": 0.21905805038335158,
+      "grad_norm": 2.3081915378570557,
+      "learning_rate": 2.5655766508755147e-05,
+      "loss": 0.9633240509033203,
+      "mean_token_accuracy": 0.741833764910698,
+      "num_tokens": 230392.0,
+      "step": 100
+    },
+    {
+      "entropy": 0.9184817910194397,
+      "epoch": 0.32858707557502737,
+      "grad_norm": 2.0965280532836914,
+      "learning_rate": 3.861322434145977e-05,
+      "loss": 0.8405924987792969,
+      "mean_token_accuracy": 0.7688284432888031,
+      "num_tokens": 341746.0,
+      "step": 150
+    },
+    {
+      "entropy": 0.8618861585855484,
+      "epoch": 0.43811610076670315,
+      "grad_norm": 2.2506542205810547,
+      "learning_rate": 5.157068217416438e-05,
+      "loss": 0.7841030883789063,
+      "mean_token_accuracy": 0.7796466761827469,
+      "num_tokens": 460483.0,
+      "step": 200
+    },
+    {
+      "entropy": 0.819869134426117,
+      "epoch": 0.547645125958379,
+      "grad_norm": 1.4058395624160767,
+      "learning_rate": 6.4528140006869e-05,
+      "loss": 0.7524736785888672,
+      "mean_token_accuracy": 0.7900535291433335,
+      "num_tokens": 570627.0,
+      "step": 250
+    },
+    {
+      "entropy": 0.8075837278366089,
+      "epoch": 0.6571741511500547,
+      "grad_norm": 1.5815014839172363,
+      "learning_rate": 7.748559783957362e-05,
+      "loss": 0.7388697814941406,
+      "mean_token_accuracy": 0.7914055424928665,
+      "num_tokens": 677217.0,
+      "step": 300
+    },
+    {
+      "entropy": 0.7868322026729584,
+      "epoch": 0.7667031763417306,
+      "grad_norm": 1.2377994060516357,
+      "learning_rate": 9.044305567227824e-05,
+      "loss": 0.7226168060302735,
+      "mean_token_accuracy": 0.7953901988267899,
+      "num_tokens": 790185.0,
+      "step": 350
+    },
+    {
+      "entropy": 0.7526325100660324,
+      "epoch": 0.8762322015334063,
+      "grad_norm": 1.203137755393982,
+      "learning_rate": 0.00010340051350498286,
+      "loss": 0.6991680908203125,
+      "mean_token_accuracy": 0.7992895567417144,
+      "num_tokens": 910080.0,
+      "step": 400
+    },
+    {
+      "entropy": 0.7626404595375061,
+      "epoch": 0.9857612267250822,
+      "grad_norm": 1.1625452041625977,
+      "learning_rate": 0.00011635797133768749,
+      "loss": 0.7055020904541016,
+      "mean_token_accuracy": 0.7997505795955658,
+      "num_tokens": 1023473.0,
+      "step": 450
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.7441067624659765,
+      "eval_loss": 0.7209385633468628,
+      "eval_mean_token_accuracy": 0.7895652093584575,
+      "eval_num_tokens": 1038250.0,
+      "eval_runtime": 45.5076,
+      "eval_samples_per_second": 22.018,
+      "eval_steps_per_second": 2.769,
+      "step": 457
+    },
+    {
+      "entropy": 0.720724637460227,
+      "epoch": 1.0941949616648412,
+      "grad_norm": 1.2537180185317993,
+      "learning_rate": 0.00011840069618947894,
+      "loss": 0.6586117553710937,
+      "mean_token_accuracy": 0.8081232917429221,
+      "num_tokens": 1128922.0,
+      "step": 500
+    },
+    {
+      "entropy": 0.7084310132265091,
+      "epoch": 1.203723986856517,
+      "grad_norm": 1.2347161769866943,
+      "learning_rate": 0.00011828501915154198,
+      "loss": 0.650611572265625,
+      "mean_token_accuracy": 0.808387525677681,
+      "num_tokens": 1245635.0,
+      "step": 550
+    },
+    {
+      "entropy": 0.706004958152771,
+      "epoch": 1.3132530120481927,
+      "grad_norm": 1.2140169143676758,
+      "learning_rate": 0.00011808319665683977,
+      "loss": 0.6442465209960937,
+      "mean_token_accuracy": 0.8122780376672745,
+      "num_tokens": 1354885.0,
+      "step": 600
+    },
+    {
+      "entropy": 0.6854291236400605,
+      "epoch": 1.4227820372398685,
+      "grad_norm": 1.1137559413909912,
+      "learning_rate": 0.00011779552303848118,
+      "loss": 0.6293138885498046,
+      "mean_token_accuracy": 0.8158071845769882,
+      "num_tokens": 1472232.0,
+      "step": 650
+    },
+    {
+      "entropy": 0.6885070586204529,
+      "epoch": 1.5323110624315444,
+      "grad_norm": 1.1074714660644531,
+      "learning_rate": 0.00011742241783280468,
+      "loss": 0.6259001541137695,
+      "mean_token_accuracy": 0.8164563828706741,
+      "num_tokens": 1587494.0,
+      "step": 700
+    },
+    {
+      "entropy": 0.6932114177942276,
+      "epoch": 1.6418400876232202,
+      "grad_norm": 1.1099495887756348,
+      "learning_rate": 0.00011696442516753647,
+      "loss": 0.6335408401489258,
+      "mean_token_accuracy": 0.8156683903932571,
+      "num_tokens": 1696927.0,
+      "step": 750
+    },
+    {
+      "entropy": 0.6787053138017655,
+      "epoch": 1.751369112814896,
+      "grad_norm": 1.2505418062210083,
+      "learning_rate": 0.00011642221296824764,
+      "loss": 0.6210909271240235,
+      "mean_token_accuracy": 0.8153067737817764,
+      "num_tokens": 1813415.0,
+      "step": 800
+    },
+    {
+      "entropy": 0.6693948286771775,
+      "epoch": 1.8608981380065717,
+      "grad_norm": 1.0963624715805054,
+      "learning_rate": 0.00011579657198426736,
+      "loss": 0.6105814361572266,
+      "mean_token_accuracy": 0.819609425663948,
+      "num_tokens": 1924911.0,
+      "step": 850
+    },
+    {
+      "entropy": 0.6634757363796234,
+      "epoch": 1.9704271631982475,
+      "grad_norm": 0.8766036033630371,
+      "learning_rate": 0.00011508841463547313,
+      "loss": 0.6100498580932617,
+      "mean_token_accuracy": 0.8197187691926956,
+      "num_tokens": 2044650.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.0,
+      "eval_entropy": 0.6342300275961558,
+      "eval_loss": 0.6517631411552429,
+      "eval_mean_token_accuracy": 0.807878240233376,
+      "eval_num_tokens": 2076500.0,
+      "eval_runtime": 44.6225,
+      "eval_samples_per_second": 22.455,
+      "eval_steps_per_second": 2.824,
+      "step": 914
+    },
+    {
+      "entropy": 0.6060671378867795,
+      "epoch": 2.078860898138007,
+      "grad_norm": 0.8028743267059326,
+      "learning_rate": 0.00011429877368163915,
+      "loss": 0.5440861511230469,
+      "mean_token_accuracy": 0.833196161973356,
+      "num_tokens": 2160498.0,
+      "step": 950
+    },
+    {
+      "entropy": 0.5839239662885666,
+      "epoch": 2.1883899233296824,
+      "grad_norm": 0.8585777878761292,
+      "learning_rate": 0.00011342880071628435,
+      "loss": 0.5182851409912109,
+      "mean_token_accuracy": 0.8406627786159515,
+      "num_tokens": 2269979.0,
+      "step": 1000
+    },
+    {
+      "entropy": 0.5949062475562096,
+      "epoch": 2.297918948521358,
+      "grad_norm": 0.9953155517578125,
+      "learning_rate": 0.00011247976448721578,
+      "loss": 0.5327045822143555,
+      "mean_token_accuracy": 0.8344444561004639,
+      "num_tokens": 2378362.0,
+      "step": 1050
+    },
+    {
+      "entropy": 0.583380132317543,
+      "epoch": 2.407447973713034,
+      "grad_norm": 0.9235143065452576,
+      "learning_rate": 0.00011145304904621704,
+      "loss": 0.520841178894043,
+      "mean_token_accuracy": 0.839934149980545,
+      "num_tokens": 2498494.0,
+      "step": 1100
+    },
+    {
+      "entropy": 0.5811308068037033,
+      "epoch": 2.5169769989047097,
+      "grad_norm": 1.1114075183868408,
+      "learning_rate": 0.00011035015173058038,
+      "loss": 0.5212027740478515,
+      "mean_token_accuracy": 0.838929146528244,
+      "num_tokens": 2611646.0,
+      "step": 1150
+    },
+    {
+      "entropy": 0.5903497302532196,
+      "epoch": 2.6265060240963853,
+      "grad_norm": 1.053895354270935,
+      "learning_rate": 0.0001091726809794255,
+      "loss": 0.5264106750488281,
+      "mean_token_accuracy": 0.8369076484441758,
+      "num_tokens": 2717744.0,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5949014449119567,
+      "epoch": 2.7360350492880614,
+      "grad_norm": 0.7563770413398743,
+      "learning_rate": 0.00010792235398799048,
+      "loss": 0.5322903442382813,
+      "mean_token_accuracy": 0.8368168956041336,
+      "num_tokens": 2827750.0,
+      "step": 1250
+    },
+    {
+      "entropy": 0.5745143675804139,
+      "epoch": 2.845564074479737,
+      "grad_norm": 0.8103823065757751,
+      "learning_rate": 0.00010660099420331526,
+      "loss": 0.5154821014404297,
+      "mean_token_accuracy": 0.8410378849506378,
+      "num_tokens": 2948958.0,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5665904894471169,
+      "epoch": 2.955093099671413,
+      "grad_norm": 0.8764206171035767,
+      "learning_rate": 0.00010521052866496986,
+      "loss": 0.5121800231933594,
+      "mean_token_accuracy": 0.8420848768949508,
+      "num_tokens": 3066917.0,
+      "step": 1350
+    },
+    {
+      "epoch": 3.0,
+      "eval_entropy": 0.5603872095308606,
+      "eval_loss": 0.6720245480537415,
+      "eval_mean_token_accuracy": 0.8079002830717299,
+      "eval_num_tokens": 3114750.0,
+      "eval_runtime": 43.9396,
+      "eval_samples_per_second": 22.804,
+      "eval_steps_per_second": 2.868,
+      "step": 1371
+    },
+    {
+      "entropy": 0.5181072778774031,
+      "epoch": 3.063526834611172,
+      "grad_norm": 0.889443039894104,
+      "learning_rate": 0.00010375298519470596,
+      "loss": 0.44869777679443357,
+      "mean_token_accuracy": 0.8574847211741438,
+      "num_tokens": 3180366.0,
+      "step": 1400
+    },
+    {
+      "entropy": 0.4902029836177826,
+      "epoch": 3.1730558598028478,
+      "grad_norm": 0.8999125957489014,
+      "learning_rate": 0.00010223048943913007,
+      "loss": 0.41646446228027345,
+      "mean_token_accuracy": 0.8655553787946701,
+      "num_tokens": 3293918.0,
+      "step": 1450
+    },
+    {
+      "entropy": 0.48207574665546415,
+      "epoch": 3.2825848849945234,
+      "grad_norm": 1.0042189359664917,
+      "learning_rate": 0.00010064526176971124,
+      "loss": 0.41205951690673825,
+      "mean_token_accuracy": 0.866786903142929,
+      "num_tokens": 3409058.0,
+      "step": 1500
+    },
+    {
+      "entropy": 0.4839185461401939,
+      "epoch": 3.3921139101861995,
+      "grad_norm": 1.1650396585464478,
+      "learning_rate": 9.89996140446443e-05,
+      "loss": 0.4102657699584961,
+      "mean_token_accuracy": 0.8668668657541275,
+      "num_tokens": 3525131.0,
+      "step": 1550
+    },
+    {
+      "entropy": 0.4932769918441772,
+      "epoch": 3.501642935377875,
+      "grad_norm": 1.0935224294662476,
+      "learning_rate": 9.729594623729099e-05,
+      "loss": 0.4225375366210937,
+      "mean_token_accuracy": 0.8634544950723648,
+      "num_tokens": 3641504.0,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5046219238638878,
+      "epoch": 3.6111719605695507,
+      "grad_norm": 0.9509316086769104,
+      "learning_rate": 9.553674293611626e-05,
+      "loss": 0.4319529342651367,
+      "mean_token_accuracy": 0.8610132175683975,
+      "num_tokens": 3752182.0,
+      "step": 1650
+    },
+    {
+      "entropy": 0.5090703725814819,
+      "epoch": 3.7207009857612268,
+      "grad_norm": 0.9050212502479553,
+      "learning_rate": 9.372456972122359e-05,
+      "loss": 0.4376397705078125,
+      "mean_token_accuracy": 0.8591927194595337,
+      "num_tokens": 3866473.0,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4903561845421791,
+      "epoch": 3.8302300109529024,
+      "grad_norm": 1.1624391078948975,
+      "learning_rate": 9.186206942277429e-05,
+      "loss": 0.4219248962402344,
+      "mean_token_accuracy": 0.8636374253034592,
+      "num_tokens": 3977393.0,
+      "step": 1750
+    },
+    {
+      "entropy": 0.5040662395954132,
+      "epoch": 3.9397590361445785,
+      "grad_norm": 0.798085629940033,
+      "learning_rate": 8.995195826674703e-05,
+      "loss": 0.4333121871948242,
+      "mean_token_accuracy": 0.86187963783741,
+      "num_tokens": 4088936.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5136857172326436,
+      "eval_loss": 0.6814945340156555,
+      "eval_mean_token_accuracy": 0.8091600262929523,
+      "eval_num_tokens": 4153000.0,
+      "eval_runtime": 44.5059,
+      "eval_samples_per_second": 22.514,
+      "eval_steps_per_second": 2.831,
+      "step": 1828
+    },
+    {
+      "entropy": 0.4410170723091472,
+      "epoch": 4.048192771084337,
+      "grad_norm": 0.9169163703918457,
+      "learning_rate": 8.799702191365879e-05,
+      "loss": 0.3653216552734375,
+      "mean_token_accuracy": 0.8799899684058295,
+      "num_tokens": 4202329.0,
+      "step": 1850
+    },
+    {
+      "entropy": 0.38991558194160464,
+      "epoch": 4.157721796276014,
+      "grad_norm": 1.2269299030303955,
+      "learning_rate": 8.600011139602396e-05,
+      "loss": 0.3035482597351074,
+      "mean_token_accuracy": 0.8981181675195694,
+      "num_tokens": 4320321.0,
+      "step": 1900
+    },
+    {
+      "entropy": 0.38429499566555025,
+      "epoch": 4.267250821467689,
+      "grad_norm": 1.1254630088806152,
+      "learning_rate": 8.39641389604766e-05,
+      "loss": 0.3037366676330566,
+      "mean_token_accuracy": 0.8979371869564057,
+      "num_tokens": 4435139.0,
+      "step": 1950
+    },
+    {
+      "entropy": 0.39086049795150757,
+      "epoch": 4.376779846659365,
+      "grad_norm": 1.3771028518676758,
+      "learning_rate": 8.18920738206195e-05,
+      "loss": 0.3100498390197754,
+      "mean_token_accuracy": 0.8953604429960251,
+      "num_tokens": 4547272.0,
+      "step": 2000
+    },
+    {
+      "entropy": 0.3967577290534973,
+      "epoch": 4.48630887185104,
+      "grad_norm": 1.0673140287399292,
+      "learning_rate": 7.978693782679394e-05,
+      "loss": 0.31265506744384763,
+      "mean_token_accuracy": 0.8946040958166123,
+      "num_tokens": 4659036.0,
+      "step": 2050
+    },
+    {
+      "entropy": 0.3927252873778343,
+      "epoch": 4.595837897042716,
+      "grad_norm": 1.1611689329147339,
+      "learning_rate": 7.765180105908515e-05,
+      "loss": 0.3075417518615723,
+      "mean_token_accuracy": 0.8964757239818573,
+      "num_tokens": 4769957.0,
+      "step": 2100
+    },
+    {
+      "entropy": 0.39732489734888077,
+      "epoch": 4.705366922234392,
+      "grad_norm": 1.3783643245697021,
+      "learning_rate": 7.548977734999099e-05,
+      "loss": 0.3155888748168945,
+      "mean_token_accuracy": 0.8941756916046143,
+      "num_tokens": 4882487.0,
+      "step": 2150
+    },
+    {
+      "entropy": 0.3972980257868767,
+      "epoch": 4.814895947426068,
+      "grad_norm": 0.9033240675926208,
+      "learning_rate": 7.330401974328283e-05,
+      "loss": 0.3149653244018555,
+      "mean_token_accuracy": 0.8941631782054901,
+      "num_tokens": 4993060.0,
+      "step": 2200
+    },
+    {
+      "entropy": 0.39401222407817843,
+      "epoch": 4.924424972617744,
+      "grad_norm": 1.1885665655136108,
+      "learning_rate": 7.109771589568177e-05,
+      "loss": 0.31299734115600586,
+      "mean_token_accuracy": 0.8955705845355988,
+      "num_tokens": 5111182.0,
+      "step": 2250
+    },
+    {
+      "epoch": 5.0,
+      "eval_entropy": 0.44022186218746123,
+      "eval_loss": 0.731863260269165,
+      "eval_mean_token_accuracy": 0.8086685883620429,
+      "eval_num_tokens": 5191250.0,
+      "eval_runtime": 44.8156,
+      "eval_samples_per_second": 22.358,
+      "eval_steps_per_second": 2.812,
+      "step": 2285
+    },
+    {
+      "entropy": 0.3665939736546892,
+      "epoch": 5.032858707557502,
+      "grad_norm": 1.2354992628097534,
+      "learning_rate": 6.887408342805626e-05,
+      "loss": 0.27955495834350585,
+      "mean_token_accuracy": 0.9059046464737015,
+      "num_tokens": 5225197.0,
+      "step": 2300
+    },
+    {
+      "entropy": 0.2878261508047581,
+      "epoch": 5.142387732749179,
+      "grad_norm": 1.3155502080917358,
+      "learning_rate": 6.663636523292057e-05,
+      "loss": 0.1942205810546875,
+      "mean_token_accuracy": 0.9343426609039307,
+      "num_tokens": 5336728.0,
+      "step": 2350
+    },
+    {
+      "entropy": 0.2881846517324448,
+      "epoch": 5.2519167579408546,
+      "grad_norm": 1.157877802848816,
+      "learning_rate": 6.438782474507771e-05,
+      "loss": 0.19725875854492186,
+      "mean_token_accuracy": 0.9327778881788253,
+      "num_tokens": 5446986.0,
+      "step": 2400
+    },
+    {
+      "entropy": 0.2804257667064667,
+      "epoch": 5.36144578313253,
+      "grad_norm": 1.3322943449020386,
+      "learning_rate": 6.213174118230412e-05,
+      "loss": 0.19409126281738281,
+      "mean_token_accuracy": 0.9335259759426117,
+      "num_tokens": 5563479.0,
+      "step": 2450
+    },
+    {
+      "entropy": 0.28692082852125167,
+      "epoch": 5.470974808324206,
+      "grad_norm": 1.2746684551239014,
+      "learning_rate": 5.987140476301658e-05,
+      "loss": 0.2002597999572754,
+      "mean_token_accuracy": 0.9313689643144607,
+      "num_tokens": 5677338.0,
+      "step": 2500
+    },
+    {
+      "entropy": 0.2895388266444206,
+      "epoch": 5.580503833515881,
+      "grad_norm": 1.149001121520996,
+      "learning_rate": 5.761011190789638e-05,
+      "loss": 0.2018396759033203,
+      "mean_token_accuracy": 0.9323018848896026,
+      "num_tokens": 5787988.0,
+      "step": 2550
+    },
+    {
+      "entropy": 0.29092250213027,
+      "epoch": 5.690032858707557,
+      "grad_norm": 1.3538148403167725,
+      "learning_rate": 5.5351160432467955e-05,
+      "loss": 0.20052331924438477,
+      "mean_token_accuracy": 0.9314265781641007,
+      "num_tokens": 5902559.0,
+      "step": 2600
+    },
+    {
+      "entropy": 0.28983849346637724,
+      "epoch": 5.7995618838992335,
+      "grad_norm": 1.2387913465499878,
+      "learning_rate": 5.309784473764352e-05,
+      "loss": 0.20004833221435547,
+      "mean_token_accuracy": 0.9308328247070312,
+      "num_tokens": 6014039.0,
+      "step": 2650
+    },
+    {
+      "entropy": 0.27550786405801775,
+      "epoch": 5.909090909090909,
+      "grad_norm": 1.1062148809432983,
+      "learning_rate": 5.085345100524733e-05,
+      "loss": 0.19400547027587892,
+      "mean_token_accuracy": 0.9349210584163665,
+      "num_tokens": 6134464.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.0,
+      "eval_entropy": 0.36818124211969827,
+      "eval_loss": 0.8581439256668091,
+      "eval_mean_token_accuracy": 0.8029990144192226,
+      "eval_num_tokens": 6229500.0,
+      "eval_runtime": 44.8208,
+      "eval_samples_per_second": 22.356,
+      "eval_steps_per_second": 2.811,
+      "step": 2742
+    },
+    {
+      "entropy": 0.2698795477850269,
+      "epoch": 6.017524644030668,
+      "grad_norm": 0.9266197085380554,
+      "learning_rate": 4.862125240552664e-05,
+      "loss": 0.18622320175170898,
+      "mean_token_accuracy": 0.9371824848531473,
+      "num_tokens": 6249905.0,
+      "step": 2750
+    },
+    {
+      "entropy": 0.19044648699462413,
+      "epoch": 6.127053669222344,
+      "grad_norm": 1.0554289817810059,
+      "learning_rate": 4.640450432363826e-05,
+      "loss": 0.11102401733398437,
+      "mean_token_accuracy": 0.9636328971385956,
+      "num_tokens": 6367665.0,
+      "step": 2800
+    },
+    {
+      "entropy": 0.1951981072127819,
+      "epoch": 6.23658269441402,
+      "grad_norm": 1.1141842603683472,
+      "learning_rate": 4.420643961207262e-05,
+      "loss": 0.11578875541687011,
+      "mean_token_accuracy": 0.9620302796363831,
+      "num_tokens": 6481769.0,
+      "step": 2850
+    },
+    {
+      "entropy": 0.20094555169343947,
+      "epoch": 6.3461117196056955,
+      "grad_norm": 0.9116718769073486,
+      "learning_rate": 4.2030263875938836e-05,
+      "loss": 0.11906105995178223,
+      "mean_token_accuracy": 0.9608682131767273,
+      "num_tokens": 6595712.0,
+      "step": 2900
+    },
+    {
+      "entropy": 0.19917954877018929,
+      "epoch": 6.455640744797371,
+      "grad_norm": 1.0561001300811768,
+      "learning_rate": 3.98791507979867e-05,
+      "loss": 0.11807474136352539,
+      "mean_token_accuracy": 0.9615647631883621,
+      "num_tokens": 6708841.0,
+      "step": 2950
+    },
+    {
+      "entropy": 0.20109083890914917,
+      "epoch": 6.565169769989047,
+      "grad_norm": 0.9614781141281128,
+      "learning_rate": 3.7756237510183554e-05,
+      "loss": 0.12305583000183105,
+      "mean_token_accuracy": 0.9595666688680649,
+      "num_tokens": 6818025.0,
+      "step": 3000
+    },
+    {
+      "entropy": 0.20449081182479859,
+      "epoch": 6.674698795180722,
+      "grad_norm": 1.1171538829803467,
+      "learning_rate": 3.566462001859587e-05,
+      "loss": 0.12225373268127442,
+      "mean_token_accuracy": 0.9595645654201508,
+      "num_tokens": 6929746.0,
+      "step": 3050
+    },
+    {
+      "entropy": 0.19336393870413304,
+      "epoch": 6.784227820372399,
+      "grad_norm": 0.9689428806304932,
+      "learning_rate": 3.3607348688248006e-05,
+      "loss": 0.11655777931213379,
+      "mean_token_accuracy": 0.9615726363658905,
+      "num_tokens": 7046109.0,
+      "step": 3100
+    },
+    {
+      "entropy": 0.20375137731432916,
+      "epoch": 6.8937568455640745,
+      "grad_norm": 1.1648379564285278,
+      "learning_rate": 3.158742379454244e-05,
+      "loss": 0.12166361808776856,
+      "mean_token_accuracy": 0.9593629771471024,
+      "num_tokens": 7156155.0,
+      "step": 3150
+    },
+    {
+      "epoch": 7.0,
+      "eval_entropy": 0.31269029941823745,
+      "eval_loss": 1.0066559314727783,
+      "eval_mean_token_accuracy": 0.802609798454103,
+      "eval_num_tokens": 7267750.0,
+      "eval_runtime": 44.8168,
+      "eval_samples_per_second": 22.358,
+      "eval_steps_per_second": 2.811,
+      "step": 3199
+    },
+    {
+      "entropy": 0.19614394039216668,
+      "epoch": 7.002190580503833,
+      "grad_norm": 0.807051420211792,
+      "learning_rate": 2.9607791147730068e-05,
+      "loss": 0.11796384811401367,
+      "mean_token_accuracy": 0.9610634036738464,
+      "num_tokens": 7269424.0,
+      "step": 3200
+    },
+    {
+      "entropy": 0.15629827834665774,
+      "epoch": 7.11171960569551,
+      "grad_norm": 0.7889683842658997,
+      "learning_rate": 2.76713377968105e-05,
+      "loss": 0.08246193885803223,
+      "mean_token_accuracy": 0.9736558765172958,
+      "num_tokens": 7382588.0,
+      "step": 3250
+    },
+    {
+      "entropy": 0.15588059544563293,
+      "epoch": 7.221248630887185,
+      "grad_norm": 0.9219930768013,
+      "learning_rate": 2.5780887819129002e-05,
+      "loss": 0.08199316024780273,
+      "mean_token_accuracy": 0.9730826634168624,
+      "num_tokens": 7494048.0,
+      "step": 3300
+    },
+    {
+      "entropy": 0.15537467963993548,
+      "epoch": 7.330777656078861,
+      "grad_norm": 0.7120053768157959,
+      "learning_rate": 2.3939198201809304e-05,
+      "loss": 0.08418251037597656,
+      "mean_token_accuracy": 0.9733713102340699,
+      "num_tokens": 7607343.0,
+      "step": 3350
+    },
+    {
+      "entropy": 0.1485396324843168,
+      "epoch": 7.4403066812705365,
+      "grad_norm": 0.5975527763366699,
+      "learning_rate": 2.2148954821029444e-05,
+      "loss": 0.07945024490356445,
+      "mean_token_accuracy": 0.9737608224153519,
+      "num_tokens": 7724520.0,
+      "step": 3400
+    },
+    {
+      "entropy": 0.15404113121330737,
+      "epoch": 7.549835706462212,
+      "grad_norm": 0.7434099912643433,
+      "learning_rate": 2.0412768525004168e-05,
+      "loss": 0.08357874870300293,
+      "mean_token_accuracy": 0.9732766461372375,
+      "num_tokens": 7836568.0,
+      "step": 3450
+    },
+    {
+      "entropy": 0.15270612128078936,
+      "epoch": 7.659364731653889,
+      "grad_norm": 0.8742797374725342,
+      "learning_rate": 1.8733171326386e-05,
+      "loss": 0.08433089256286622,
+      "mean_token_accuracy": 0.9724701988697052,
+      "num_tokens": 7947988.0,
+      "step": 3500
+    },
+    {
+      "entropy": 0.148405454531312,
+      "epoch": 7.768893756845564,
+      "grad_norm": 0.6974013447761536,
+      "learning_rate": 1.711261270963867e-05,
+      "loss": 0.07989848613739013,
+      "mean_token_accuracy": 0.9736948871612549,
+      "num_tokens": 8066809.0,
+      "step": 3550
+    },
+    {
+      "entropy": 0.1545567201822996,
+      "epoch": 7.87842278203724,
+      "grad_norm": 0.6734868288040161,
+      "learning_rate": 1.5553456058767216e-05,
+      "loss": 0.0851497745513916,
+      "mean_token_accuracy": 0.9718592113256455,
+      "num_tokens": 8178429.0,
+      "step": 3600
+    },
+    {
+      "entropy": 0.1492402908951044,
+      "epoch": 7.9879518072289155,
+      "grad_norm": 0.6337016224861145,
+      "learning_rate": 1.4057975210615318e-05,
+      "loss": 0.08205368041992188,
+      "mean_token_accuracy": 0.9729280340671539,
+      "num_tokens": 8294498.0,
+      "step": 3650
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2715017907912769,
+      "eval_loss": 1.1209602355957031,
+      "eval_mean_token_accuracy": 0.8031659982507191,
+      "eval_num_tokens": 8306000.0,
+      "eval_runtime": 44.5068,
+      "eval_samples_per_second": 22.513,
+      "eval_steps_per_second": 2.831,
+      "step": 3656
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 4570,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.8007844674315584e+17,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4113/README.md b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4113/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..7ac50e2706e5b4eae8f28e204601afb241c49e55
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4113/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-2B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-2B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4113/adapter_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4113/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..86d878f780798dd8f06b10d3bdb6eaee702b2da7
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4113/adapter_config.json
@@ -0,0 +1,48 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.01686122009693236,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "v_proj",
+    "o_proj",
+    "k_proj",
+    "up_proj",
+    "q_proj",
+    "down_proj",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4113/chat_template.jinja b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4113/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..0ef09f214eaa6d9bca297988afc1454b5827b2c7
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4113/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is true %}
+        {{- '<think>\n' }}
+    {%- else %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4113/tokenizer_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4113/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4113/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4113/trainer_state.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4113/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..f6b5a9607dcb79099fa749071ecd7b398e6741a7
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4113/trainer_state.json
@@ -0,0 +1,953 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.0,
+  "eval_steps": 500,
+  "global_step": 4113,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.074948468208313,
+      "epoch": 0.10952902519167579,
+      "grad_norm": 2.3511903285980225,
+      "learning_rate": 1.2698308676050528e-05,
+      "loss": 1.9153688049316406,
+      "mean_token_accuracy": 0.6077924159169197,
+      "num_tokens": 113087.0,
+      "step": 50
+    },
+    {
+      "entropy": 1.054056368470192,
+      "epoch": 0.21905805038335158,
+      "grad_norm": 2.3081915378570557,
+      "learning_rate": 2.5655766508755147e-05,
+      "loss": 0.9633240509033203,
+      "mean_token_accuracy": 0.741833764910698,
+      "num_tokens": 230392.0,
+      "step": 100
+    },
+    {
+      "entropy": 0.9184817910194397,
+      "epoch": 0.32858707557502737,
+      "grad_norm": 2.0965280532836914,
+      "learning_rate": 3.861322434145977e-05,
+      "loss": 0.8405924987792969,
+      "mean_token_accuracy": 0.7688284432888031,
+      "num_tokens": 341746.0,
+      "step": 150
+    },
+    {
+      "entropy": 0.8618861585855484,
+      "epoch": 0.43811610076670315,
+      "grad_norm": 2.2506542205810547,
+      "learning_rate": 5.157068217416438e-05,
+      "loss": 0.7841030883789063,
+      "mean_token_accuracy": 0.7796466761827469,
+      "num_tokens": 460483.0,
+      "step": 200
+    },
+    {
+      "entropy": 0.819869134426117,
+      "epoch": 0.547645125958379,
+      "grad_norm": 1.4058395624160767,
+      "learning_rate": 6.4528140006869e-05,
+      "loss": 0.7524736785888672,
+      "mean_token_accuracy": 0.7900535291433335,
+      "num_tokens": 570627.0,
+      "step": 250
+    },
+    {
+      "entropy": 0.8075837278366089,
+      "epoch": 0.6571741511500547,
+      "grad_norm": 1.5815014839172363,
+      "learning_rate": 7.748559783957362e-05,
+      "loss": 0.7388697814941406,
+      "mean_token_accuracy": 0.7914055424928665,
+      "num_tokens": 677217.0,
+      "step": 300
+    },
+    {
+      "entropy": 0.7868322026729584,
+      "epoch": 0.7667031763417306,
+      "grad_norm": 1.2377994060516357,
+      "learning_rate": 9.044305567227824e-05,
+      "loss": 0.7226168060302735,
+      "mean_token_accuracy": 0.7953901988267899,
+      "num_tokens": 790185.0,
+      "step": 350
+    },
+    {
+      "entropy": 0.7526325100660324,
+      "epoch": 0.8762322015334063,
+      "grad_norm": 1.203137755393982,
+      "learning_rate": 0.00010340051350498286,
+      "loss": 0.6991680908203125,
+      "mean_token_accuracy": 0.7992895567417144,
+      "num_tokens": 910080.0,
+      "step": 400
+    },
+    {
+      "entropy": 0.7626404595375061,
+      "epoch": 0.9857612267250822,
+      "grad_norm": 1.1625452041625977,
+      "learning_rate": 0.00011635797133768749,
+      "loss": 0.7055020904541016,
+      "mean_token_accuracy": 0.7997505795955658,
+      "num_tokens": 1023473.0,
+      "step": 450
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.7441067624659765,
+      "eval_loss": 0.7209385633468628,
+      "eval_mean_token_accuracy": 0.7895652093584575,
+      "eval_num_tokens": 1038250.0,
+      "eval_runtime": 45.5076,
+      "eval_samples_per_second": 22.018,
+      "eval_steps_per_second": 2.769,
+      "step": 457
+    },
+    {
+      "entropy": 0.720724637460227,
+      "epoch": 1.0941949616648412,
+      "grad_norm": 1.2537180185317993,
+      "learning_rate": 0.00011840069618947894,
+      "loss": 0.6586117553710937,
+      "mean_token_accuracy": 0.8081232917429221,
+      "num_tokens": 1128922.0,
+      "step": 500
+    },
+    {
+      "entropy": 0.7084310132265091,
+      "epoch": 1.203723986856517,
+      "grad_norm": 1.2347161769866943,
+      "learning_rate": 0.00011828501915154198,
+      "loss": 0.650611572265625,
+      "mean_token_accuracy": 0.808387525677681,
+      "num_tokens": 1245635.0,
+      "step": 550
+    },
+    {
+      "entropy": 0.706004958152771,
+      "epoch": 1.3132530120481927,
+      "grad_norm": 1.2140169143676758,
+      "learning_rate": 0.00011808319665683977,
+      "loss": 0.6442465209960937,
+      "mean_token_accuracy": 0.8122780376672745,
+      "num_tokens": 1354885.0,
+      "step": 600
+    },
+    {
+      "entropy": 0.6854291236400605,
+      "epoch": 1.4227820372398685,
+      "grad_norm": 1.1137559413909912,
+      "learning_rate": 0.00011779552303848118,
+      "loss": 0.6293138885498046,
+      "mean_token_accuracy": 0.8158071845769882,
+      "num_tokens": 1472232.0,
+      "step": 650
+    },
+    {
+      "entropy": 0.6885070586204529,
+      "epoch": 1.5323110624315444,
+      "grad_norm": 1.1074714660644531,
+      "learning_rate": 0.00011742241783280468,
+      "loss": 0.6259001541137695,
+      "mean_token_accuracy": 0.8164563828706741,
+      "num_tokens": 1587494.0,
+      "step": 700
+    },
+    {
+      "entropy": 0.6932114177942276,
+      "epoch": 1.6418400876232202,
+      "grad_norm": 1.1099495887756348,
+      "learning_rate": 0.00011696442516753647,
+      "loss": 0.6335408401489258,
+      "mean_token_accuracy": 0.8156683903932571,
+      "num_tokens": 1696927.0,
+      "step": 750
+    },
+    {
+      "entropy": 0.6787053138017655,
+      "epoch": 1.751369112814896,
+      "grad_norm": 1.2505418062210083,
+      "learning_rate": 0.00011642221296824764,
+      "loss": 0.6210909271240235,
+      "mean_token_accuracy": 0.8153067737817764,
+      "num_tokens": 1813415.0,
+      "step": 800
+    },
+    {
+      "entropy": 0.6693948286771775,
+      "epoch": 1.8608981380065717,
+      "grad_norm": 1.0963624715805054,
+      "learning_rate": 0.00011579657198426736,
+      "loss": 0.6105814361572266,
+      "mean_token_accuracy": 0.819609425663948,
+      "num_tokens": 1924911.0,
+      "step": 850
+    },
+    {
+      "entropy": 0.6634757363796234,
+      "epoch": 1.9704271631982475,
+      "grad_norm": 0.8766036033630371,
+      "learning_rate": 0.00011508841463547313,
+      "loss": 0.6100498580932617,
+      "mean_token_accuracy": 0.8197187691926956,
+      "num_tokens": 2044650.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.0,
+      "eval_entropy": 0.6342300275961558,
+      "eval_loss": 0.6517631411552429,
+      "eval_mean_token_accuracy": 0.807878240233376,
+      "eval_num_tokens": 2076500.0,
+      "eval_runtime": 44.6225,
+      "eval_samples_per_second": 22.455,
+      "eval_steps_per_second": 2.824,
+      "step": 914
+    },
+    {
+      "entropy": 0.6060671378867795,
+      "epoch": 2.078860898138007,
+      "grad_norm": 0.8028743267059326,
+      "learning_rate": 0.00011429877368163915,
+      "loss": 0.5440861511230469,
+      "mean_token_accuracy": 0.833196161973356,
+      "num_tokens": 2160498.0,
+      "step": 950
+    },
+    {
+      "entropy": 0.5839239662885666,
+      "epoch": 2.1883899233296824,
+      "grad_norm": 0.8585777878761292,
+      "learning_rate": 0.00011342880071628435,
+      "loss": 0.5182851409912109,
+      "mean_token_accuracy": 0.8406627786159515,
+      "num_tokens": 2269979.0,
+      "step": 1000
+    },
+    {
+      "entropy": 0.5949062475562096,
+      "epoch": 2.297918948521358,
+      "grad_norm": 0.9953155517578125,
+      "learning_rate": 0.00011247976448721578,
+      "loss": 0.5327045822143555,
+      "mean_token_accuracy": 0.8344444561004639,
+      "num_tokens": 2378362.0,
+      "step": 1050
+    },
+    {
+      "entropy": 0.583380132317543,
+      "epoch": 2.407447973713034,
+      "grad_norm": 0.9235143065452576,
+      "learning_rate": 0.00011145304904621704,
+      "loss": 0.520841178894043,
+      "mean_token_accuracy": 0.839934149980545,
+      "num_tokens": 2498494.0,
+      "step": 1100
+    },
+    {
+      "entropy": 0.5811308068037033,
+      "epoch": 2.5169769989047097,
+      "grad_norm": 1.1114075183868408,
+      "learning_rate": 0.00011035015173058038,
+      "loss": 0.5212027740478515,
+      "mean_token_accuracy": 0.838929146528244,
+      "num_tokens": 2611646.0,
+      "step": 1150
+    },
+    {
+      "entropy": 0.5903497302532196,
+      "epoch": 2.6265060240963853,
+      "grad_norm": 1.053895354270935,
+      "learning_rate": 0.0001091726809794255,
+      "loss": 0.5264106750488281,
+      "mean_token_accuracy": 0.8369076484441758,
+      "num_tokens": 2717744.0,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5949014449119567,
+      "epoch": 2.7360350492880614,
+      "grad_norm": 0.7563770413398743,
+      "learning_rate": 0.00010792235398799048,
+      "loss": 0.5322903442382813,
+      "mean_token_accuracy": 0.8368168956041336,
+      "num_tokens": 2827750.0,
+      "step": 1250
+    },
+    {
+      "entropy": 0.5745143675804139,
+      "epoch": 2.845564074479737,
+      "grad_norm": 0.8103823065757751,
+      "learning_rate": 0.00010660099420331526,
+      "loss": 0.5154821014404297,
+      "mean_token_accuracy": 0.8410378849506378,
+      "num_tokens": 2948958.0,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5665904894471169,
+      "epoch": 2.955093099671413,
+      "grad_norm": 0.8764206171035767,
+      "learning_rate": 0.00010521052866496986,
+      "loss": 0.5121800231933594,
+      "mean_token_accuracy": 0.8420848768949508,
+      "num_tokens": 3066917.0,
+      "step": 1350
+    },
+    {
+      "epoch": 3.0,
+      "eval_entropy": 0.5603872095308606,
+      "eval_loss": 0.6720245480537415,
+      "eval_mean_token_accuracy": 0.8079002830717299,
+      "eval_num_tokens": 3114750.0,
+      "eval_runtime": 43.9396,
+      "eval_samples_per_second": 22.804,
+      "eval_steps_per_second": 2.868,
+      "step": 1371
+    },
+    {
+      "entropy": 0.5181072778774031,
+      "epoch": 3.063526834611172,
+      "grad_norm": 0.889443039894104,
+      "learning_rate": 0.00010375298519470596,
+      "loss": 0.44869777679443357,
+      "mean_token_accuracy": 0.8574847211741438,
+      "num_tokens": 3180366.0,
+      "step": 1400
+    },
+    {
+      "entropy": 0.4902029836177826,
+      "epoch": 3.1730558598028478,
+      "grad_norm": 0.8999125957489014,
+      "learning_rate": 0.00010223048943913007,
+      "loss": 0.41646446228027345,
+      "mean_token_accuracy": 0.8655553787946701,
+      "num_tokens": 3293918.0,
+      "step": 1450
+    },
+    {
+      "entropy": 0.48207574665546415,
+      "epoch": 3.2825848849945234,
+      "grad_norm": 1.0042189359664917,
+      "learning_rate": 0.00010064526176971124,
+      "loss": 0.41205951690673825,
+      "mean_token_accuracy": 0.866786903142929,
+      "num_tokens": 3409058.0,
+      "step": 1500
+    },
+    {
+      "entropy": 0.4839185461401939,
+      "epoch": 3.3921139101861995,
+      "grad_norm": 1.1650396585464478,
+      "learning_rate": 9.89996140446443e-05,
+      "loss": 0.4102657699584961,
+      "mean_token_accuracy": 0.8668668657541275,
+      "num_tokens": 3525131.0,
+      "step": 1550
+    },
+    {
+      "entropy": 0.4932769918441772,
+      "epoch": 3.501642935377875,
+      "grad_norm": 1.0935224294662476,
+      "learning_rate": 9.729594623729099e-05,
+      "loss": 0.4225375366210937,
+      "mean_token_accuracy": 0.8634544950723648,
+      "num_tokens": 3641504.0,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5046219238638878,
+      "epoch": 3.6111719605695507,
+      "grad_norm": 0.9509316086769104,
+      "learning_rate": 9.553674293611626e-05,
+      "loss": 0.4319529342651367,
+      "mean_token_accuracy": 0.8610132175683975,
+      "num_tokens": 3752182.0,
+      "step": 1650
+    },
+    {
+      "entropy": 0.5090703725814819,
+      "epoch": 3.7207009857612268,
+      "grad_norm": 0.9050212502479553,
+      "learning_rate": 9.372456972122359e-05,
+      "loss": 0.4376397705078125,
+      "mean_token_accuracy": 0.8591927194595337,
+      "num_tokens": 3866473.0,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4903561845421791,
+      "epoch": 3.8302300109529024,
+      "grad_norm": 1.1624391078948975,
+      "learning_rate": 9.186206942277429e-05,
+      "loss": 0.4219248962402344,
+      "mean_token_accuracy": 0.8636374253034592,
+      "num_tokens": 3977393.0,
+      "step": 1750
+    },
+    {
+      "entropy": 0.5040662395954132,
+      "epoch": 3.9397590361445785,
+      "grad_norm": 0.798085629940033,
+      "learning_rate": 8.995195826674703e-05,
+      "loss": 0.4333121871948242,
+      "mean_token_accuracy": 0.86187963783741,
+      "num_tokens": 4088936.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5136857172326436,
+      "eval_loss": 0.6814945340156555,
+      "eval_mean_token_accuracy": 0.8091600262929523,
+      "eval_num_tokens": 4153000.0,
+      "eval_runtime": 44.5059,
+      "eval_samples_per_second": 22.514,
+      "eval_steps_per_second": 2.831,
+      "step": 1828
+    },
+    {
+      "entropy": 0.4410170723091472,
+      "epoch": 4.048192771084337,
+      "grad_norm": 0.9169163703918457,
+      "learning_rate": 8.799702191365879e-05,
+      "loss": 0.3653216552734375,
+      "mean_token_accuracy": 0.8799899684058295,
+      "num_tokens": 4202329.0,
+      "step": 1850
+    },
+    {
+      "entropy": 0.38991558194160464,
+      "epoch": 4.157721796276014,
+      "grad_norm": 1.2269299030303955,
+      "learning_rate": 8.600011139602396e-05,
+      "loss": 0.3035482597351074,
+      "mean_token_accuracy": 0.8981181675195694,
+      "num_tokens": 4320321.0,
+      "step": 1900
+    },
+    {
+      "entropy": 0.38429499566555025,
+      "epoch": 4.267250821467689,
+      "grad_norm": 1.1254630088806152,
+      "learning_rate": 8.39641389604766e-05,
+      "loss": 0.3037366676330566,
+      "mean_token_accuracy": 0.8979371869564057,
+      "num_tokens": 4435139.0,
+      "step": 1950
+    },
+    {
+      "entropy": 0.39086049795150757,
+      "epoch": 4.376779846659365,
+      "grad_norm": 1.3771028518676758,
+      "learning_rate": 8.18920738206195e-05,
+      "loss": 0.3100498390197754,
+      "mean_token_accuracy": 0.8953604429960251,
+      "num_tokens": 4547272.0,
+      "step": 2000
+    },
+    {
+      "entropy": 0.3967577290534973,
+      "epoch": 4.48630887185104,
+      "grad_norm": 1.0673140287399292,
+      "learning_rate": 7.978693782679394e-05,
+      "loss": 0.31265506744384763,
+      "mean_token_accuracy": 0.8946040958166123,
+      "num_tokens": 4659036.0,
+      "step": 2050
+    },
+    {
+      "entropy": 0.3927252873778343,
+      "epoch": 4.595837897042716,
+      "grad_norm": 1.1611689329147339,
+      "learning_rate": 7.765180105908515e-05,
+      "loss": 0.3075417518615723,
+      "mean_token_accuracy": 0.8964757239818573,
+      "num_tokens": 4769957.0,
+      "step": 2100
+    },
+    {
+      "entropy": 0.39732489734888077,
+      "epoch": 4.705366922234392,
+      "grad_norm": 1.3783643245697021,
+      "learning_rate": 7.548977734999099e-05,
+      "loss": 0.3155888748168945,
+      "mean_token_accuracy": 0.8941756916046143,
+      "num_tokens": 4882487.0,
+      "step": 2150
+    },
+    {
+      "entropy": 0.3972980257868767,
+      "epoch": 4.814895947426068,
+      "grad_norm": 0.9033240675926208,
+      "learning_rate": 7.330401974328283e-05,
+      "loss": 0.3149653244018555,
+      "mean_token_accuracy": 0.8941631782054901,
+      "num_tokens": 4993060.0,
+      "step": 2200
+    },
+    {
+      "entropy": 0.39401222407817843,
+      "epoch": 4.924424972617744,
+      "grad_norm": 1.1885665655136108,
+      "learning_rate": 7.109771589568177e-05,
+      "loss": 0.31299734115600586,
+      "mean_token_accuracy": 0.8955705845355988,
+      "num_tokens": 5111182.0,
+      "step": 2250
+    },
+    {
+      "epoch": 5.0,
+      "eval_entropy": 0.44022186218746123,
+      "eval_loss": 0.731863260269165,
+      "eval_mean_token_accuracy": 0.8086685883620429,
+      "eval_num_tokens": 5191250.0,
+      "eval_runtime": 44.8156,
+      "eval_samples_per_second": 22.358,
+      "eval_steps_per_second": 2.812,
+      "step": 2285
+    },
+    {
+      "entropy": 0.3665939736546892,
+      "epoch": 5.032858707557502,
+      "grad_norm": 1.2354992628097534,
+      "learning_rate": 6.887408342805626e-05,
+      "loss": 0.27955495834350585,
+      "mean_token_accuracy": 0.9059046464737015,
+      "num_tokens": 5225197.0,
+      "step": 2300
+    },
+    {
+      "entropy": 0.2878261508047581,
+      "epoch": 5.142387732749179,
+      "grad_norm": 1.3155502080917358,
+      "learning_rate": 6.663636523292057e-05,
+      "loss": 0.1942205810546875,
+      "mean_token_accuracy": 0.9343426609039307,
+      "num_tokens": 5336728.0,
+      "step": 2350
+    },
+    {
+      "entropy": 0.2881846517324448,
+      "epoch": 5.2519167579408546,
+      "grad_norm": 1.157877802848816,
+      "learning_rate": 6.438782474507771e-05,
+      "loss": 0.19725875854492186,
+      "mean_token_accuracy": 0.9327778881788253,
+      "num_tokens": 5446986.0,
+      "step": 2400
+    },
+    {
+      "entropy": 0.2804257667064667,
+      "epoch": 5.36144578313253,
+      "grad_norm": 1.3322943449020386,
+      "learning_rate": 6.213174118230412e-05,
+      "loss": 0.19409126281738281,
+      "mean_token_accuracy": 0.9335259759426117,
+      "num_tokens": 5563479.0,
+      "step": 2450
+    },
+    {
+      "entropy": 0.28692082852125167,
+      "epoch": 5.470974808324206,
+      "grad_norm": 1.2746684551239014,
+      "learning_rate": 5.987140476301658e-05,
+      "loss": 0.2002597999572754,
+      "mean_token_accuracy": 0.9313689643144607,
+      "num_tokens": 5677338.0,
+      "step": 2500
+    },
+    {
+      "entropy": 0.2895388266444206,
+      "epoch": 5.580503833515881,
+      "grad_norm": 1.149001121520996,
+      "learning_rate": 5.761011190789638e-05,
+      "loss": 0.2018396759033203,
+      "mean_token_accuracy": 0.9323018848896026,
+      "num_tokens": 5787988.0,
+      "step": 2550
+    },
+    {
+      "entropy": 0.29092250213027,
+      "epoch": 5.690032858707557,
+      "grad_norm": 1.3538148403167725,
+      "learning_rate": 5.5351160432467955e-05,
+      "loss": 0.20052331924438477,
+      "mean_token_accuracy": 0.9314265781641007,
+      "num_tokens": 5902559.0,
+      "step": 2600
+    },
+    {
+      "entropy": 0.28983849346637724,
+      "epoch": 5.7995618838992335,
+      "grad_norm": 1.2387913465499878,
+      "learning_rate": 5.309784473764352e-05,
+      "loss": 0.20004833221435547,
+      "mean_token_accuracy": 0.9308328247070312,
+      "num_tokens": 6014039.0,
+      "step": 2650
+    },
+    {
+      "entropy": 0.27550786405801775,
+      "epoch": 5.909090909090909,
+      "grad_norm": 1.1062148809432983,
+      "learning_rate": 5.085345100524733e-05,
+      "loss": 0.19400547027587892,
+      "mean_token_accuracy": 0.9349210584163665,
+      "num_tokens": 6134464.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.0,
+      "eval_entropy": 0.36818124211969827,
+      "eval_loss": 0.8581439256668091,
+      "eval_mean_token_accuracy": 0.8029990144192226,
+      "eval_num_tokens": 6229500.0,
+      "eval_runtime": 44.8208,
+      "eval_samples_per_second": 22.356,
+      "eval_steps_per_second": 2.811,
+      "step": 2742
+    },
+    {
+      "entropy": 0.2698795477850269,
+      "epoch": 6.017524644030668,
+      "grad_norm": 0.9266197085380554,
+      "learning_rate": 4.862125240552664e-05,
+      "loss": 0.18622320175170898,
+      "mean_token_accuracy": 0.9371824848531473,
+      "num_tokens": 6249905.0,
+      "step": 2750
+    },
+    {
+      "entropy": 0.19044648699462413,
+      "epoch": 6.127053669222344,
+      "grad_norm": 1.0554289817810059,
+      "learning_rate": 4.640450432363826e-05,
+      "loss": 0.11102401733398437,
+      "mean_token_accuracy": 0.9636328971385956,
+      "num_tokens": 6367665.0,
+      "step": 2800
+    },
+    {
+      "entropy": 0.1951981072127819,
+      "epoch": 6.23658269441402,
+      "grad_norm": 1.1141842603683472,
+      "learning_rate": 4.420643961207262e-05,
+      "loss": 0.11578875541687011,
+      "mean_token_accuracy": 0.9620302796363831,
+      "num_tokens": 6481769.0,
+      "step": 2850
+    },
+    {
+      "entropy": 0.20094555169343947,
+      "epoch": 6.3461117196056955,
+      "grad_norm": 0.9116718769073486,
+      "learning_rate": 4.2030263875938836e-05,
+      "loss": 0.11906105995178223,
+      "mean_token_accuracy": 0.9608682131767273,
+      "num_tokens": 6595712.0,
+      "step": 2900
+    },
+    {
+      "entropy": 0.19917954877018929,
+      "epoch": 6.455640744797371,
+      "grad_norm": 1.0561001300811768,
+      "learning_rate": 3.98791507979867e-05,
+      "loss": 0.11807474136352539,
+      "mean_token_accuracy": 0.9615647631883621,
+      "num_tokens": 6708841.0,
+      "step": 2950
+    },
+    {
+      "entropy": 0.20109083890914917,
+      "epoch": 6.565169769989047,
+      "grad_norm": 0.9614781141281128,
+      "learning_rate": 3.7756237510183554e-05,
+      "loss": 0.12305583000183105,
+      "mean_token_accuracy": 0.9595666688680649,
+      "num_tokens": 6818025.0,
+      "step": 3000
+    },
+    {
+      "entropy": 0.20449081182479859,
+      "epoch": 6.674698795180722,
+      "grad_norm": 1.1171538829803467,
+      "learning_rate": 3.566462001859587e-05,
+      "loss": 0.12225373268127442,
+      "mean_token_accuracy": 0.9595645654201508,
+      "num_tokens": 6929746.0,
+      "step": 3050
+    },
+    {
+      "entropy": 0.19336393870413304,
+      "epoch": 6.784227820372399,
+      "grad_norm": 0.9689428806304932,
+      "learning_rate": 3.3607348688248006e-05,
+      "loss": 0.11655777931213379,
+      "mean_token_accuracy": 0.9615726363658905,
+      "num_tokens": 7046109.0,
+      "step": 3100
+    },
+    {
+      "entropy": 0.20375137731432916,
+      "epoch": 6.8937568455640745,
+      "grad_norm": 1.1648379564285278,
+      "learning_rate": 3.158742379454244e-05,
+      "loss": 0.12166361808776856,
+      "mean_token_accuracy": 0.9593629771471024,
+      "num_tokens": 7156155.0,
+      "step": 3150
+    },
+    {
+      "epoch": 7.0,
+      "eval_entropy": 0.31269029941823745,
+      "eval_loss": 1.0066559314727783,
+      "eval_mean_token_accuracy": 0.802609798454103,
+      "eval_num_tokens": 7267750.0,
+      "eval_runtime": 44.8168,
+      "eval_samples_per_second": 22.358,
+      "eval_steps_per_second": 2.811,
+      "step": 3199
+    },
+    {
+      "entropy": 0.19614394039216668,
+      "epoch": 7.002190580503833,
+      "grad_norm": 0.807051420211792,
+      "learning_rate": 2.9607791147730068e-05,
+      "loss": 0.11796384811401367,
+      "mean_token_accuracy": 0.9610634036738464,
+      "num_tokens": 7269424.0,
+      "step": 3200
+    },
+    {
+      "entropy": 0.15629827834665774,
+      "epoch": 7.11171960569551,
+      "grad_norm": 0.7889683842658997,
+      "learning_rate": 2.76713377968105e-05,
+      "loss": 0.08246193885803223,
+      "mean_token_accuracy": 0.9736558765172958,
+      "num_tokens": 7382588.0,
+      "step": 3250
+    },
+    {
+      "entropy": 0.15588059544563293,
+      "epoch": 7.221248630887185,
+      "grad_norm": 0.9219930768013,
+      "learning_rate": 2.5780887819129002e-05,
+      "loss": 0.08199316024780273,
+      "mean_token_accuracy": 0.9730826634168624,
+      "num_tokens": 7494048.0,
+      "step": 3300
+    },
+    {
+      "entropy": 0.15537467963993548,
+      "epoch": 7.330777656078861,
+      "grad_norm": 0.7120053768157959,
+      "learning_rate": 2.3939198201809304e-05,
+      "loss": 0.08418251037597656,
+      "mean_token_accuracy": 0.9733713102340699,
+      "num_tokens": 7607343.0,
+      "step": 3350
+    },
+    {
+      "entropy": 0.1485396324843168,
+      "epoch": 7.4403066812705365,
+      "grad_norm": 0.5975527763366699,
+      "learning_rate": 2.2148954821029444e-05,
+      "loss": 0.07945024490356445,
+      "mean_token_accuracy": 0.9737608224153519,
+      "num_tokens": 7724520.0,
+      "step": 3400
+    },
+    {
+      "entropy": 0.15404113121330737,
+      "epoch": 7.549835706462212,
+      "grad_norm": 0.7434099912643433,
+      "learning_rate": 2.0412768525004168e-05,
+      "loss": 0.08357874870300293,
+      "mean_token_accuracy": 0.9732766461372375,
+      "num_tokens": 7836568.0,
+      "step": 3450
+    },
+    {
+      "entropy": 0.15270612128078936,
+      "epoch": 7.659364731653889,
+      "grad_norm": 0.8742797374725342,
+      "learning_rate": 1.8733171326386e-05,
+      "loss": 0.08433089256286622,
+      "mean_token_accuracy": 0.9724701988697052,
+      "num_tokens": 7947988.0,
+      "step": 3500
+    },
+    {
+      "entropy": 0.148405454531312,
+      "epoch": 7.768893756845564,
+      "grad_norm": 0.6974013447761536,
+      "learning_rate": 1.711261270963867e-05,
+      "loss": 0.07989848613739013,
+      "mean_token_accuracy": 0.9736948871612549,
+      "num_tokens": 8066809.0,
+      "step": 3550
+    },
+    {
+      "entropy": 0.1545567201822996,
+      "epoch": 7.87842278203724,
+      "grad_norm": 0.6734868288040161,
+      "learning_rate": 1.5553456058767216e-05,
+      "loss": 0.0851497745513916,
+      "mean_token_accuracy": 0.9718592113256455,
+      "num_tokens": 8178429.0,
+      "step": 3600
+    },
+    {
+      "entropy": 0.1492402908951044,
+      "epoch": 7.9879518072289155,
+      "grad_norm": 0.6337016224861145,
+      "learning_rate": 1.4057975210615318e-05,
+      "loss": 0.08205368041992188,
+      "mean_token_accuracy": 0.9729280340671539,
+      "num_tokens": 8294498.0,
+      "step": 3650
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2715017907912769,
+      "eval_loss": 1.1209602355957031,
+      "eval_mean_token_accuracy": 0.8031659982507191,
+      "eval_num_tokens": 8306000.0,
+      "eval_runtime": 44.5068,
+      "eval_samples_per_second": 22.513,
+      "eval_steps_per_second": 2.831,
+      "step": 3656
+    },
+    {
+      "entropy": 0.13691872885130874,
+      "epoch": 8.096385542168674,
+      "grad_norm": 0.46566903591156006,
+      "learning_rate": 1.2628351138755825e-05,
+      "loss": 0.06957323551177978,
+      "mean_token_accuracy": 0.9769269468808415,
+      "num_tokens": 8408606.0,
+      "step": 3700
+    },
+    {
+      "entropy": 0.13172941066324712,
+      "epoch": 8.20591456736035,
+      "grad_norm": 0.5115252137184143,
+      "learning_rate": 1.1266668772811068e-05,
+      "loss": 0.06850167751312256,
+      "mean_token_accuracy": 0.9768519496917725,
+      "num_tokens": 8523042.0,
+      "step": 3750
+    },
+    {
+      "entropy": 0.12640717133879661,
+      "epoch": 8.315443592552027,
+      "grad_norm": 0.419202983379364,
+      "learning_rate": 9.974913957841114e-06,
+      "loss": 0.0659257173538208,
+      "mean_token_accuracy": 0.9778822070360184,
+      "num_tokens": 8640478.0,
+      "step": 3800
+    },
+    {
+      "entropy": 0.1372934626042843,
+      "epoch": 8.424972617743702,
+      "grad_norm": 0.6855530738830566,
+      "learning_rate": 8.754970558234856e-06,
+      "loss": 0.07134664535522461,
+      "mean_token_accuracy": 0.9757242006063461,
+      "num_tokens": 8752683.0,
+      "step": 3850
+    },
+    {
+      "entropy": 0.13225064136087894,
+      "epoch": 8.534501642935378,
+      "grad_norm": 0.40154290199279785,
+      "learning_rate": 7.6086177103270694e-06,
+      "loss": 0.0688913869857788,
+      "mean_token_accuracy": 0.9764530754089356,
+      "num_tokens": 8869555.0,
+      "step": 3900
+    },
+    {
+      "entropy": 0.14314913783222438,
+      "epoch": 8.644030668127053,
+      "grad_norm": 0.6097934246063232,
+      "learning_rate": 6.537527227748551e-06,
+      "loss": 0.07303566455841065,
+      "mean_token_accuracy": 0.9746374058723449,
+      "num_tokens": 8977869.0,
+      "step": 3950
+    },
+    {
+      "entropy": 0.13993739932775498,
+      "epoch": 8.75355969331873,
+      "grad_norm": 0.3800851106643677,
+      "learning_rate": 5.543261163292889e-06,
+      "loss": 0.07268502235412598,
+      "mean_token_accuracy": 0.9749921530485153,
+      "num_tokens": 9085599.0,
+      "step": 4000
+    },
+    {
+      "entropy": 0.13490288957953453,
+      "epoch": 8.863088718510404,
+      "grad_norm": 0.4685533940792084,
+      "learning_rate": 4.627269530856084e-06,
+      "loss": 0.07146972179412842,
+      "mean_token_accuracy": 0.9757514011859894,
+      "num_tokens": 9196726.0,
+      "step": 4050
+    },
+    {
+      "entropy": 0.129932743832469,
+      "epoch": 8.97261774370208,
+      "grad_norm": 0.4819929003715515,
+      "learning_rate": 3.7908881907707993e-06,
+      "loss": 0.06786080360412598,
+      "mean_token_accuracy": 0.9767654579877854,
+      "num_tokens": 9314678.0,
+      "step": 4100
+    },
+    {
+      "epoch": 9.0,
+      "eval_entropy": 0.2552800567613708,
+      "eval_loss": 1.2286313772201538,
+      "eval_mean_token_accuracy": 0.8038436865049695,
+      "eval_num_tokens": 9344250.0,
+      "eval_runtime": 44.7709,
+      "eval_samples_per_second": 22.381,
+      "eval_steps_per_second": 2.814,
+      "step": 4113
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 4570,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.0262519999913536e+17,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-457/README.md b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-457/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..7ac50e2706e5b4eae8f28e204601afb241c49e55
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-457/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-2B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-2B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-457/adapter_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-457/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..86d878f780798dd8f06b10d3bdb6eaee702b2da7
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-457/adapter_config.json
@@ -0,0 +1,48 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.01686122009693236,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "v_proj",
+    "o_proj",
+    "k_proj",
+    "up_proj",
+    "q_proj",
+    "down_proj",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-457/chat_template.jinja b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-457/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..0ef09f214eaa6d9bca297988afc1454b5827b2c7
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-457/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is true %}
+        {{- '<think>\n' }}
+    {%- else %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-457/tokenizer_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-457/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-457/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-457/trainer_state.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-457/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..acffd44a26986e59f8e0693d7b58911ab7e783f9
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-457/trainer_state.json
@@ -0,0 +1,135 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "eval_steps": 500,
+  "global_step": 457,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.074948468208313,
+      "epoch": 0.10952902519167579,
+      "grad_norm": 2.3511903285980225,
+      "learning_rate": 1.2698308676050528e-05,
+      "loss": 1.9153688049316406,
+      "mean_token_accuracy": 0.6077924159169197,
+      "num_tokens": 113087.0,
+      "step": 50
+    },
+    {
+      "entropy": 1.054056368470192,
+      "epoch": 0.21905805038335158,
+      "grad_norm": 2.3081915378570557,
+      "learning_rate": 2.5655766508755147e-05,
+      "loss": 0.9633240509033203,
+      "mean_token_accuracy": 0.741833764910698,
+      "num_tokens": 230392.0,
+      "step": 100
+    },
+    {
+      "entropy": 0.9184817910194397,
+      "epoch": 0.32858707557502737,
+      "grad_norm": 2.0965280532836914,
+      "learning_rate": 3.861322434145977e-05,
+      "loss": 0.8405924987792969,
+      "mean_token_accuracy": 0.7688284432888031,
+      "num_tokens": 341746.0,
+      "step": 150
+    },
+    {
+      "entropy": 0.8618861585855484,
+      "epoch": 0.43811610076670315,
+      "grad_norm": 2.2506542205810547,
+      "learning_rate": 5.157068217416438e-05,
+      "loss": 0.7841030883789063,
+      "mean_token_accuracy": 0.7796466761827469,
+      "num_tokens": 460483.0,
+      "step": 200
+    },
+    {
+      "entropy": 0.819869134426117,
+      "epoch": 0.547645125958379,
+      "grad_norm": 1.4058395624160767,
+      "learning_rate": 6.4528140006869e-05,
+      "loss": 0.7524736785888672,
+      "mean_token_accuracy": 0.7900535291433335,
+      "num_tokens": 570627.0,
+      "step": 250
+    },
+    {
+      "entropy": 0.8075837278366089,
+      "epoch": 0.6571741511500547,
+      "grad_norm": 1.5815014839172363,
+      "learning_rate": 7.748559783957362e-05,
+      "loss": 0.7388697814941406,
+      "mean_token_accuracy": 0.7914055424928665,
+      "num_tokens": 677217.0,
+      "step": 300
+    },
+    {
+      "entropy": 0.7868322026729584,
+      "epoch": 0.7667031763417306,
+      "grad_norm": 1.2377994060516357,
+      "learning_rate": 9.044305567227824e-05,
+      "loss": 0.7226168060302735,
+      "mean_token_accuracy": 0.7953901988267899,
+      "num_tokens": 790185.0,
+      "step": 350
+    },
+    {
+      "entropy": 0.7526325100660324,
+      "epoch": 0.8762322015334063,
+      "grad_norm": 1.203137755393982,
+      "learning_rate": 0.00010340051350498286,
+      "loss": 0.6991680908203125,
+      "mean_token_accuracy": 0.7992895567417144,
+      "num_tokens": 910080.0,
+      "step": 400
+    },
+    {
+      "entropy": 0.7626404595375061,
+      "epoch": 0.9857612267250822,
+      "grad_norm": 1.1625452041625977,
+      "learning_rate": 0.00011635797133768749,
+      "loss": 0.7055020904541016,
+      "mean_token_accuracy": 0.7997505795955658,
+      "num_tokens": 1023473.0,
+      "step": 450
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.7441067624659765,
+      "eval_loss": 0.7209385633468628,
+      "eval_mean_token_accuracy": 0.7895652093584575,
+      "eval_num_tokens": 1038250.0,
+      "eval_runtime": 45.5076,
+      "eval_samples_per_second": 22.018,
+      "eval_steps_per_second": 2.769,
+      "step": 457
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 4570,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.245682367536256e+16,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4570/README.md b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4570/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..7ac50e2706e5b4eae8f28e204601afb241c49e55
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4570/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-2B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-2B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4570/adapter_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4570/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..86d878f780798dd8f06b10d3bdb6eaee702b2da7
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4570/adapter_config.json
@@ -0,0 +1,48 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.01686122009693236,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "v_proj",
+    "o_proj",
+    "k_proj",
+    "up_proj",
+    "q_proj",
+    "down_proj",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4570/chat_template.jinja b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4570/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..0ef09f214eaa6d9bca297988afc1454b5827b2c7
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4570/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is true %}
+        {{- '<think>\n' }}
+    {%- else %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4570/tokenizer_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4570/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4570/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4570/trainer_state.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4570/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..80f37ea44b8c4e7d2a806f840ce92fa1f0476d36
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-4570/trainer_state.json
@@ -0,0 +1,1054 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 10.0,
+  "eval_steps": 500,
+  "global_step": 4570,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 2.074948468208313,
+      "epoch": 0.10952902519167579,
+      "grad_norm": 2.3511903285980225,
+      "learning_rate": 1.2698308676050528e-05,
+      "loss": 1.9153688049316406,
+      "mean_token_accuracy": 0.6077924159169197,
+      "num_tokens": 113087.0,
+      "step": 50
+    },
+    {
+      "entropy": 1.054056368470192,
+      "epoch": 0.21905805038335158,
+      "grad_norm": 2.3081915378570557,
+      "learning_rate": 2.5655766508755147e-05,
+      "loss": 0.9633240509033203,
+      "mean_token_accuracy": 0.741833764910698,
+      "num_tokens": 230392.0,
+      "step": 100
+    },
+    {
+      "entropy": 0.9184817910194397,
+      "epoch": 0.32858707557502737,
+      "grad_norm": 2.0965280532836914,
+      "learning_rate": 3.861322434145977e-05,
+      "loss": 0.8405924987792969,
+      "mean_token_accuracy": 0.7688284432888031,
+      "num_tokens": 341746.0,
+      "step": 150
+    },
+    {
+      "entropy": 0.8618861585855484,
+      "epoch": 0.43811610076670315,
+      "grad_norm": 2.2506542205810547,
+      "learning_rate": 5.157068217416438e-05,
+      "loss": 0.7841030883789063,
+      "mean_token_accuracy": 0.7796466761827469,
+      "num_tokens": 460483.0,
+      "step": 200
+    },
+    {
+      "entropy": 0.819869134426117,
+      "epoch": 0.547645125958379,
+      "grad_norm": 1.4058395624160767,
+      "learning_rate": 6.4528140006869e-05,
+      "loss": 0.7524736785888672,
+      "mean_token_accuracy": 0.7900535291433335,
+      "num_tokens": 570627.0,
+      "step": 250
+    },
+    {
+      "entropy": 0.8075837278366089,
+      "epoch": 0.6571741511500547,
+      "grad_norm": 1.5815014839172363,
+      "learning_rate": 7.748559783957362e-05,
+      "loss": 0.7388697814941406,
+      "mean_token_accuracy": 0.7914055424928665,
+      "num_tokens": 677217.0,
+      "step": 300
+    },
+    {
+      "entropy": 0.7868322026729584,
+      "epoch": 0.7667031763417306,
+      "grad_norm": 1.2377994060516357,
+      "learning_rate": 9.044305567227824e-05,
+      "loss": 0.7226168060302735,
+      "mean_token_accuracy": 0.7953901988267899,
+      "num_tokens": 790185.0,
+      "step": 350
+    },
+    {
+      "entropy": 0.7526325100660324,
+      "epoch": 0.8762322015334063,
+      "grad_norm": 1.203137755393982,
+      "learning_rate": 0.00010340051350498286,
+      "loss": 0.6991680908203125,
+      "mean_token_accuracy": 0.7992895567417144,
+      "num_tokens": 910080.0,
+      "step": 400
+    },
+    {
+      "entropy": 0.7626404595375061,
+      "epoch": 0.9857612267250822,
+      "grad_norm": 1.1625452041625977,
+      "learning_rate": 0.00011635797133768749,
+      "loss": 0.7055020904541016,
+      "mean_token_accuracy": 0.7997505795955658,
+      "num_tokens": 1023473.0,
+      "step": 450
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.7441067624659765,
+      "eval_loss": 0.7209385633468628,
+      "eval_mean_token_accuracy": 0.7895652093584575,
+      "eval_num_tokens": 1038250.0,
+      "eval_runtime": 45.5076,
+      "eval_samples_per_second": 22.018,
+      "eval_steps_per_second": 2.769,
+      "step": 457
+    },
+    {
+      "entropy": 0.720724637460227,
+      "epoch": 1.0941949616648412,
+      "grad_norm": 1.2537180185317993,
+      "learning_rate": 0.00011840069618947894,
+      "loss": 0.6586117553710937,
+      "mean_token_accuracy": 0.8081232917429221,
+      "num_tokens": 1128922.0,
+      "step": 500
+    },
+    {
+      "entropy": 0.7084310132265091,
+      "epoch": 1.203723986856517,
+      "grad_norm": 1.2347161769866943,
+      "learning_rate": 0.00011828501915154198,
+      "loss": 0.650611572265625,
+      "mean_token_accuracy": 0.808387525677681,
+      "num_tokens": 1245635.0,
+      "step": 550
+    },
+    {
+      "entropy": 0.706004958152771,
+      "epoch": 1.3132530120481927,
+      "grad_norm": 1.2140169143676758,
+      "learning_rate": 0.00011808319665683977,
+      "loss": 0.6442465209960937,
+      "mean_token_accuracy": 0.8122780376672745,
+      "num_tokens": 1354885.0,
+      "step": 600
+    },
+    {
+      "entropy": 0.6854291236400605,
+      "epoch": 1.4227820372398685,
+      "grad_norm": 1.1137559413909912,
+      "learning_rate": 0.00011779552303848118,
+      "loss": 0.6293138885498046,
+      "mean_token_accuracy": 0.8158071845769882,
+      "num_tokens": 1472232.0,
+      "step": 650
+    },
+    {
+      "entropy": 0.6885070586204529,
+      "epoch": 1.5323110624315444,
+      "grad_norm": 1.1074714660644531,
+      "learning_rate": 0.00011742241783280468,
+      "loss": 0.6259001541137695,
+      "mean_token_accuracy": 0.8164563828706741,
+      "num_tokens": 1587494.0,
+      "step": 700
+    },
+    {
+      "entropy": 0.6932114177942276,
+      "epoch": 1.6418400876232202,
+      "grad_norm": 1.1099495887756348,
+      "learning_rate": 0.00011696442516753647,
+      "loss": 0.6335408401489258,
+      "mean_token_accuracy": 0.8156683903932571,
+      "num_tokens": 1696927.0,
+      "step": 750
+    },
+    {
+      "entropy": 0.6787053138017655,
+      "epoch": 1.751369112814896,
+      "grad_norm": 1.2505418062210083,
+      "learning_rate": 0.00011642221296824764,
+      "loss": 0.6210909271240235,
+      "mean_token_accuracy": 0.8153067737817764,
+      "num_tokens": 1813415.0,
+      "step": 800
+    },
+    {
+      "entropy": 0.6693948286771775,
+      "epoch": 1.8608981380065717,
+      "grad_norm": 1.0963624715805054,
+      "learning_rate": 0.00011579657198426736,
+      "loss": 0.6105814361572266,
+      "mean_token_accuracy": 0.819609425663948,
+      "num_tokens": 1924911.0,
+      "step": 850
+    },
+    {
+      "entropy": 0.6634757363796234,
+      "epoch": 1.9704271631982475,
+      "grad_norm": 0.8766036033630371,
+      "learning_rate": 0.00011508841463547313,
+      "loss": 0.6100498580932617,
+      "mean_token_accuracy": 0.8197187691926956,
+      "num_tokens": 2044650.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.0,
+      "eval_entropy": 0.6342300275961558,
+      "eval_loss": 0.6517631411552429,
+      "eval_mean_token_accuracy": 0.807878240233376,
+      "eval_num_tokens": 2076500.0,
+      "eval_runtime": 44.6225,
+      "eval_samples_per_second": 22.455,
+      "eval_steps_per_second": 2.824,
+      "step": 914
+    },
+    {
+      "entropy": 0.6060671378867795,
+      "epoch": 2.078860898138007,
+      "grad_norm": 0.8028743267059326,
+      "learning_rate": 0.00011429877368163915,
+      "loss": 0.5440861511230469,
+      "mean_token_accuracy": 0.833196161973356,
+      "num_tokens": 2160498.0,
+      "step": 950
+    },
+    {
+      "entropy": 0.5839239662885666,
+      "epoch": 2.1883899233296824,
+      "grad_norm": 0.8585777878761292,
+      "learning_rate": 0.00011342880071628435,
+      "loss": 0.5182851409912109,
+      "mean_token_accuracy": 0.8406627786159515,
+      "num_tokens": 2269979.0,
+      "step": 1000
+    },
+    {
+      "entropy": 0.5949062475562096,
+      "epoch": 2.297918948521358,
+      "grad_norm": 0.9953155517578125,
+      "learning_rate": 0.00011247976448721578,
+      "loss": 0.5327045822143555,
+      "mean_token_accuracy": 0.8344444561004639,
+      "num_tokens": 2378362.0,
+      "step": 1050
+    },
+    {
+      "entropy": 0.583380132317543,
+      "epoch": 2.407447973713034,
+      "grad_norm": 0.9235143065452576,
+      "learning_rate": 0.00011145304904621704,
+      "loss": 0.520841178894043,
+      "mean_token_accuracy": 0.839934149980545,
+      "num_tokens": 2498494.0,
+      "step": 1100
+    },
+    {
+      "entropy": 0.5811308068037033,
+      "epoch": 2.5169769989047097,
+      "grad_norm": 1.1114075183868408,
+      "learning_rate": 0.00011035015173058038,
+      "loss": 0.5212027740478515,
+      "mean_token_accuracy": 0.838929146528244,
+      "num_tokens": 2611646.0,
+      "step": 1150
+    },
+    {
+      "entropy": 0.5903497302532196,
+      "epoch": 2.6265060240963853,
+      "grad_norm": 1.053895354270935,
+      "learning_rate": 0.0001091726809794255,
+      "loss": 0.5264106750488281,
+      "mean_token_accuracy": 0.8369076484441758,
+      "num_tokens": 2717744.0,
+      "step": 1200
+    },
+    {
+      "entropy": 0.5949014449119567,
+      "epoch": 2.7360350492880614,
+      "grad_norm": 0.7563770413398743,
+      "learning_rate": 0.00010792235398799048,
+      "loss": 0.5322903442382813,
+      "mean_token_accuracy": 0.8368168956041336,
+      "num_tokens": 2827750.0,
+      "step": 1250
+    },
+    {
+      "entropy": 0.5745143675804139,
+      "epoch": 2.845564074479737,
+      "grad_norm": 0.8103823065757751,
+      "learning_rate": 0.00010660099420331526,
+      "loss": 0.5154821014404297,
+      "mean_token_accuracy": 0.8410378849506378,
+      "num_tokens": 2948958.0,
+      "step": 1300
+    },
+    {
+      "entropy": 0.5665904894471169,
+      "epoch": 2.955093099671413,
+      "grad_norm": 0.8764206171035767,
+      "learning_rate": 0.00010521052866496986,
+      "loss": 0.5121800231933594,
+      "mean_token_accuracy": 0.8420848768949508,
+      "num_tokens": 3066917.0,
+      "step": 1350
+    },
+    {
+      "epoch": 3.0,
+      "eval_entropy": 0.5603872095308606,
+      "eval_loss": 0.6720245480537415,
+      "eval_mean_token_accuracy": 0.8079002830717299,
+      "eval_num_tokens": 3114750.0,
+      "eval_runtime": 43.9396,
+      "eval_samples_per_second": 22.804,
+      "eval_steps_per_second": 2.868,
+      "step": 1371
+    },
+    {
+      "entropy": 0.5181072778774031,
+      "epoch": 3.063526834611172,
+      "grad_norm": 0.889443039894104,
+      "learning_rate": 0.00010375298519470596,
+      "loss": 0.44869777679443357,
+      "mean_token_accuracy": 0.8574847211741438,
+      "num_tokens": 3180366.0,
+      "step": 1400
+    },
+    {
+      "entropy": 0.4902029836177826,
+      "epoch": 3.1730558598028478,
+      "grad_norm": 0.8999125957489014,
+      "learning_rate": 0.00010223048943913007,
+      "loss": 0.41646446228027345,
+      "mean_token_accuracy": 0.8655553787946701,
+      "num_tokens": 3293918.0,
+      "step": 1450
+    },
+    {
+      "entropy": 0.48207574665546415,
+      "epoch": 3.2825848849945234,
+      "grad_norm": 1.0042189359664917,
+      "learning_rate": 0.00010064526176971124,
+      "loss": 0.41205951690673825,
+      "mean_token_accuracy": 0.866786903142929,
+      "num_tokens": 3409058.0,
+      "step": 1500
+    },
+    {
+      "entropy": 0.4839185461401939,
+      "epoch": 3.3921139101861995,
+      "grad_norm": 1.1650396585464478,
+      "learning_rate": 9.89996140446443e-05,
+      "loss": 0.4102657699584961,
+      "mean_token_accuracy": 0.8668668657541275,
+      "num_tokens": 3525131.0,
+      "step": 1550
+    },
+    {
+      "entropy": 0.4932769918441772,
+      "epoch": 3.501642935377875,
+      "grad_norm": 1.0935224294662476,
+      "learning_rate": 9.729594623729099e-05,
+      "loss": 0.4225375366210937,
+      "mean_token_accuracy": 0.8634544950723648,
+      "num_tokens": 3641504.0,
+      "step": 1600
+    },
+    {
+      "entropy": 0.5046219238638878,
+      "epoch": 3.6111719605695507,
+      "grad_norm": 0.9509316086769104,
+      "learning_rate": 9.553674293611626e-05,
+      "loss": 0.4319529342651367,
+      "mean_token_accuracy": 0.8610132175683975,
+      "num_tokens": 3752182.0,
+      "step": 1650
+    },
+    {
+      "entropy": 0.5090703725814819,
+      "epoch": 3.7207009857612268,
+      "grad_norm": 0.9050212502479553,
+      "learning_rate": 9.372456972122359e-05,
+      "loss": 0.4376397705078125,
+      "mean_token_accuracy": 0.8591927194595337,
+      "num_tokens": 3866473.0,
+      "step": 1700
+    },
+    {
+      "entropy": 0.4903561845421791,
+      "epoch": 3.8302300109529024,
+      "grad_norm": 1.1624391078948975,
+      "learning_rate": 9.186206942277429e-05,
+      "loss": 0.4219248962402344,
+      "mean_token_accuracy": 0.8636374253034592,
+      "num_tokens": 3977393.0,
+      "step": 1750
+    },
+    {
+      "entropy": 0.5040662395954132,
+      "epoch": 3.9397590361445785,
+      "grad_norm": 0.798085629940033,
+      "learning_rate": 8.995195826674703e-05,
+      "loss": 0.4333121871948242,
+      "mean_token_accuracy": 0.86187963783741,
+      "num_tokens": 4088936.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5136857172326436,
+      "eval_loss": 0.6814945340156555,
+      "eval_mean_token_accuracy": 0.8091600262929523,
+      "eval_num_tokens": 4153000.0,
+      "eval_runtime": 44.5059,
+      "eval_samples_per_second": 22.514,
+      "eval_steps_per_second": 2.831,
+      "step": 1828
+    },
+    {
+      "entropy": 0.4410170723091472,
+      "epoch": 4.048192771084337,
+      "grad_norm": 0.9169163703918457,
+      "learning_rate": 8.799702191365879e-05,
+      "loss": 0.3653216552734375,
+      "mean_token_accuracy": 0.8799899684058295,
+      "num_tokens": 4202329.0,
+      "step": 1850
+    },
+    {
+      "entropy": 0.38991558194160464,
+      "epoch": 4.157721796276014,
+      "grad_norm": 1.2269299030303955,
+      "learning_rate": 8.600011139602396e-05,
+      "loss": 0.3035482597351074,
+      "mean_token_accuracy": 0.8981181675195694,
+      "num_tokens": 4320321.0,
+      "step": 1900
+    },
+    {
+      "entropy": 0.38429499566555025,
+      "epoch": 4.267250821467689,
+      "grad_norm": 1.1254630088806152,
+      "learning_rate": 8.39641389604766e-05,
+      "loss": 0.3037366676330566,
+      "mean_token_accuracy": 0.8979371869564057,
+      "num_tokens": 4435139.0,
+      "step": 1950
+    },
+    {
+      "entropy": 0.39086049795150757,
+      "epoch": 4.376779846659365,
+      "grad_norm": 1.3771028518676758,
+      "learning_rate": 8.18920738206195e-05,
+      "loss": 0.3100498390197754,
+      "mean_token_accuracy": 0.8953604429960251,
+      "num_tokens": 4547272.0,
+      "step": 2000
+    },
+    {
+      "entropy": 0.3967577290534973,
+      "epoch": 4.48630887185104,
+      "grad_norm": 1.0673140287399292,
+      "learning_rate": 7.978693782679394e-05,
+      "loss": 0.31265506744384763,
+      "mean_token_accuracy": 0.8946040958166123,
+      "num_tokens": 4659036.0,
+      "step": 2050
+    },
+    {
+      "entropy": 0.3927252873778343,
+      "epoch": 4.595837897042716,
+      "grad_norm": 1.1611689329147339,
+      "learning_rate": 7.765180105908515e-05,
+      "loss": 0.3075417518615723,
+      "mean_token_accuracy": 0.8964757239818573,
+      "num_tokens": 4769957.0,
+      "step": 2100
+    },
+    {
+      "entropy": 0.39732489734888077,
+      "epoch": 4.705366922234392,
+      "grad_norm": 1.3783643245697021,
+      "learning_rate": 7.548977734999099e-05,
+      "loss": 0.3155888748168945,
+      "mean_token_accuracy": 0.8941756916046143,
+      "num_tokens": 4882487.0,
+      "step": 2150
+    },
+    {
+      "entropy": 0.3972980257868767,
+      "epoch": 4.814895947426068,
+      "grad_norm": 0.9033240675926208,
+      "learning_rate": 7.330401974328283e-05,
+      "loss": 0.3149653244018555,
+      "mean_token_accuracy": 0.8941631782054901,
+      "num_tokens": 4993060.0,
+      "step": 2200
+    },
+    {
+      "entropy": 0.39401222407817843,
+      "epoch": 4.924424972617744,
+      "grad_norm": 1.1885665655136108,
+      "learning_rate": 7.109771589568177e-05,
+      "loss": 0.31299734115600586,
+      "mean_token_accuracy": 0.8955705845355988,
+      "num_tokens": 5111182.0,
+      "step": 2250
+    },
+    {
+      "epoch": 5.0,
+      "eval_entropy": 0.44022186218746123,
+      "eval_loss": 0.731863260269165,
+      "eval_mean_token_accuracy": 0.8086685883620429,
+      "eval_num_tokens": 5191250.0,
+      "eval_runtime": 44.8156,
+      "eval_samples_per_second": 22.358,
+      "eval_steps_per_second": 2.812,
+      "step": 2285
+    },
+    {
+      "entropy": 0.3665939736546892,
+      "epoch": 5.032858707557502,
+      "grad_norm": 1.2354992628097534,
+      "learning_rate": 6.887408342805626e-05,
+      "loss": 0.27955495834350585,
+      "mean_token_accuracy": 0.9059046464737015,
+      "num_tokens": 5225197.0,
+      "step": 2300
+    },
+    {
+      "entropy": 0.2878261508047581,
+      "epoch": 5.142387732749179,
+      "grad_norm": 1.3155502080917358,
+      "learning_rate": 6.663636523292057e-05,
+      "loss": 0.1942205810546875,
+      "mean_token_accuracy": 0.9343426609039307,
+      "num_tokens": 5336728.0,
+      "step": 2350
+    },
+    {
+      "entropy": 0.2881846517324448,
+      "epoch": 5.2519167579408546,
+      "grad_norm": 1.157877802848816,
+      "learning_rate": 6.438782474507771e-05,
+      "loss": 0.19725875854492186,
+      "mean_token_accuracy": 0.9327778881788253,
+      "num_tokens": 5446986.0,
+      "step": 2400
+    },
+    {
+      "entropy": 0.2804257667064667,
+      "epoch": 5.36144578313253,
+      "grad_norm": 1.3322943449020386,
+      "learning_rate": 6.213174118230412e-05,
+      "loss": 0.19409126281738281,
+      "mean_token_accuracy": 0.9335259759426117,
+      "num_tokens": 5563479.0,
+      "step": 2450
+    },
+    {
+      "entropy": 0.28692082852125167,
+      "epoch": 5.470974808324206,
+      "grad_norm": 1.2746684551239014,
+      "learning_rate": 5.987140476301658e-05,
+      "loss": 0.2002597999572754,
+      "mean_token_accuracy": 0.9313689643144607,
+      "num_tokens": 5677338.0,
+      "step": 2500
+    },
+    {
+      "entropy": 0.2895388266444206,
+      "epoch": 5.580503833515881,
+      "grad_norm": 1.149001121520996,
+      "learning_rate": 5.761011190789638e-05,
+      "loss": 0.2018396759033203,
+      "mean_token_accuracy": 0.9323018848896026,
+      "num_tokens": 5787988.0,
+      "step": 2550
+    },
+    {
+      "entropy": 0.29092250213027,
+      "epoch": 5.690032858707557,
+      "grad_norm": 1.3538148403167725,
+      "learning_rate": 5.5351160432467955e-05,
+      "loss": 0.20052331924438477,
+      "mean_token_accuracy": 0.9314265781641007,
+      "num_tokens": 5902559.0,
+      "step": 2600
+    },
+    {
+      "entropy": 0.28983849346637724,
+      "epoch": 5.7995618838992335,
+      "grad_norm": 1.2387913465499878,
+      "learning_rate": 5.309784473764352e-05,
+      "loss": 0.20004833221435547,
+      "mean_token_accuracy": 0.9308328247070312,
+      "num_tokens": 6014039.0,
+      "step": 2650
+    },
+    {
+      "entropy": 0.27550786405801775,
+      "epoch": 5.909090909090909,
+      "grad_norm": 1.1062148809432983,
+      "learning_rate": 5.085345100524733e-05,
+      "loss": 0.19400547027587892,
+      "mean_token_accuracy": 0.9349210584163665,
+      "num_tokens": 6134464.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.0,
+      "eval_entropy": 0.36818124211969827,
+      "eval_loss": 0.8581439256668091,
+      "eval_mean_token_accuracy": 0.8029990144192226,
+      "eval_num_tokens": 6229500.0,
+      "eval_runtime": 44.8208,
+      "eval_samples_per_second": 22.356,
+      "eval_steps_per_second": 2.811,
+      "step": 2742
+    },
+    {
+      "entropy": 0.2698795477850269,
+      "epoch": 6.017524644030668,
+      "grad_norm": 0.9266197085380554,
+      "learning_rate": 4.862125240552664e-05,
+      "loss": 0.18622320175170898,
+      "mean_token_accuracy": 0.9371824848531473,
+      "num_tokens": 6249905.0,
+      "step": 2750
+    },
+    {
+      "entropy": 0.19044648699462413,
+      "epoch": 6.127053669222344,
+      "grad_norm": 1.0554289817810059,
+      "learning_rate": 4.640450432363826e-05,
+      "loss": 0.11102401733398437,
+      "mean_token_accuracy": 0.9636328971385956,
+      "num_tokens": 6367665.0,
+      "step": 2800
+    },
+    {
+      "entropy": 0.1951981072127819,
+      "epoch": 6.23658269441402,
+      "grad_norm": 1.1141842603683472,
+      "learning_rate": 4.420643961207262e-05,
+      "loss": 0.11578875541687011,
+      "mean_token_accuracy": 0.9620302796363831,
+      "num_tokens": 6481769.0,
+      "step": 2850
+    },
+    {
+      "entropy": 0.20094555169343947,
+      "epoch": 6.3461117196056955,
+      "grad_norm": 0.9116718769073486,
+      "learning_rate": 4.2030263875938836e-05,
+      "loss": 0.11906105995178223,
+      "mean_token_accuracy": 0.9608682131767273,
+      "num_tokens": 6595712.0,
+      "step": 2900
+    },
+    {
+      "entropy": 0.19917954877018929,
+      "epoch": 6.455640744797371,
+      "grad_norm": 1.0561001300811768,
+      "learning_rate": 3.98791507979867e-05,
+      "loss": 0.11807474136352539,
+      "mean_token_accuracy": 0.9615647631883621,
+      "num_tokens": 6708841.0,
+      "step": 2950
+    },
+    {
+      "entropy": 0.20109083890914917,
+      "epoch": 6.565169769989047,
+      "grad_norm": 0.9614781141281128,
+      "learning_rate": 3.7756237510183554e-05,
+      "loss": 0.12305583000183105,
+      "mean_token_accuracy": 0.9595666688680649,
+      "num_tokens": 6818025.0,
+      "step": 3000
+    },
+    {
+      "entropy": 0.20449081182479859,
+      "epoch": 6.674698795180722,
+      "grad_norm": 1.1171538829803467,
+      "learning_rate": 3.566462001859587e-05,
+      "loss": 0.12225373268127442,
+      "mean_token_accuracy": 0.9595645654201508,
+      "num_tokens": 6929746.0,
+      "step": 3050
+    },
+    {
+      "entropy": 0.19336393870413304,
+      "epoch": 6.784227820372399,
+      "grad_norm": 0.9689428806304932,
+      "learning_rate": 3.3607348688248006e-05,
+      "loss": 0.11655777931213379,
+      "mean_token_accuracy": 0.9615726363658905,
+      "num_tokens": 7046109.0,
+      "step": 3100
+    },
+    {
+      "entropy": 0.20375137731432916,
+      "epoch": 6.8937568455640745,
+      "grad_norm": 1.1648379564285278,
+      "learning_rate": 3.158742379454244e-05,
+      "loss": 0.12166361808776856,
+      "mean_token_accuracy": 0.9593629771471024,
+      "num_tokens": 7156155.0,
+      "step": 3150
+    },
+    {
+      "epoch": 7.0,
+      "eval_entropy": 0.31269029941823745,
+      "eval_loss": 1.0066559314727783,
+      "eval_mean_token_accuracy": 0.802609798454103,
+      "eval_num_tokens": 7267750.0,
+      "eval_runtime": 44.8168,
+      "eval_samples_per_second": 22.358,
+      "eval_steps_per_second": 2.811,
+      "step": 3199
+    },
+    {
+      "entropy": 0.19614394039216668,
+      "epoch": 7.002190580503833,
+      "grad_norm": 0.807051420211792,
+      "learning_rate": 2.9607791147730068e-05,
+      "loss": 0.11796384811401367,
+      "mean_token_accuracy": 0.9610634036738464,
+      "num_tokens": 7269424.0,
+      "step": 3200
+    },
+    {
+      "entropy": 0.15629827834665774,
+      "epoch": 7.11171960569551,
+      "grad_norm": 0.7889683842658997,
+      "learning_rate": 2.76713377968105e-05,
+      "loss": 0.08246193885803223,
+      "mean_token_accuracy": 0.9736558765172958,
+      "num_tokens": 7382588.0,
+      "step": 3250
+    },
+    {
+      "entropy": 0.15588059544563293,
+      "epoch": 7.221248630887185,
+      "grad_norm": 0.9219930768013,
+      "learning_rate": 2.5780887819129002e-05,
+      "loss": 0.08199316024780273,
+      "mean_token_accuracy": 0.9730826634168624,
+      "num_tokens": 7494048.0,
+      "step": 3300
+    },
+    {
+      "entropy": 0.15537467963993548,
+      "epoch": 7.330777656078861,
+      "grad_norm": 0.7120053768157959,
+      "learning_rate": 2.3939198201809304e-05,
+      "loss": 0.08418251037597656,
+      "mean_token_accuracy": 0.9733713102340699,
+      "num_tokens": 7607343.0,
+      "step": 3350
+    },
+    {
+      "entropy": 0.1485396324843168,
+      "epoch": 7.4403066812705365,
+      "grad_norm": 0.5975527763366699,
+      "learning_rate": 2.2148954821029444e-05,
+      "loss": 0.07945024490356445,
+      "mean_token_accuracy": 0.9737608224153519,
+      "num_tokens": 7724520.0,
+      "step": 3400
+    },
+    {
+      "entropy": 0.15404113121330737,
+      "epoch": 7.549835706462212,
+      "grad_norm": 0.7434099912643433,
+      "learning_rate": 2.0412768525004168e-05,
+      "loss": 0.08357874870300293,
+      "mean_token_accuracy": 0.9732766461372375,
+      "num_tokens": 7836568.0,
+      "step": 3450
+    },
+    {
+      "entropy": 0.15270612128078936,
+      "epoch": 7.659364731653889,
+      "grad_norm": 0.8742797374725342,
+      "learning_rate": 1.8733171326386e-05,
+      "loss": 0.08433089256286622,
+      "mean_token_accuracy": 0.9724701988697052,
+      "num_tokens": 7947988.0,
+      "step": 3500
+    },
+    {
+      "entropy": 0.148405454531312,
+      "epoch": 7.768893756845564,
+      "grad_norm": 0.6974013447761536,
+      "learning_rate": 1.711261270963867e-05,
+      "loss": 0.07989848613739013,
+      "mean_token_accuracy": 0.9736948871612549,
+      "num_tokens": 8066809.0,
+      "step": 3550
+    },
+    {
+      "entropy": 0.1545567201822996,
+      "epoch": 7.87842278203724,
+      "grad_norm": 0.6734868288040161,
+      "learning_rate": 1.5553456058767216e-05,
+      "loss": 0.0851497745513916,
+      "mean_token_accuracy": 0.9718592113256455,
+      "num_tokens": 8178429.0,
+      "step": 3600
+    },
+    {
+      "entropy": 0.1492402908951044,
+      "epoch": 7.9879518072289155,
+      "grad_norm": 0.6337016224861145,
+      "learning_rate": 1.4057975210615318e-05,
+      "loss": 0.08205368041992188,
+      "mean_token_accuracy": 0.9729280340671539,
+      "num_tokens": 8294498.0,
+      "step": 3650
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.2715017907912769,
+      "eval_loss": 1.1209602355957031,
+      "eval_mean_token_accuracy": 0.8031659982507191,
+      "eval_num_tokens": 8306000.0,
+      "eval_runtime": 44.5068,
+      "eval_samples_per_second": 22.513,
+      "eval_steps_per_second": 2.831,
+      "step": 3656
+    },
+    {
+      "entropy": 0.13691872885130874,
+      "epoch": 8.096385542168674,
+      "grad_norm": 0.46566903591156006,
+      "learning_rate": 1.2628351138755825e-05,
+      "loss": 0.06957323551177978,
+      "mean_token_accuracy": 0.9769269468808415,
+      "num_tokens": 8408606.0,
+      "step": 3700
+    },
+    {
+      "entropy": 0.13172941066324712,
+      "epoch": 8.20591456736035,
+      "grad_norm": 0.5115252137184143,
+      "learning_rate": 1.1266668772811068e-05,
+      "loss": 0.06850167751312256,
+      "mean_token_accuracy": 0.9768519496917725,
+      "num_tokens": 8523042.0,
+      "step": 3750
+    },
+    {
+      "entropy": 0.12640717133879661,
+      "epoch": 8.315443592552027,
+      "grad_norm": 0.419202983379364,
+      "learning_rate": 9.974913957841114e-06,
+      "loss": 0.0659257173538208,
+      "mean_token_accuracy": 0.9778822070360184,
+      "num_tokens": 8640478.0,
+      "step": 3800
+    },
+    {
+      "entropy": 0.1372934626042843,
+      "epoch": 8.424972617743702,
+      "grad_norm": 0.6855530738830566,
+      "learning_rate": 8.754970558234856e-06,
+      "loss": 0.07134664535522461,
+      "mean_token_accuracy": 0.9757242006063461,
+      "num_tokens": 8752683.0,
+      "step": 3850
+    },
+    {
+      "entropy": 0.13225064136087894,
+      "epoch": 8.534501642935378,
+      "grad_norm": 0.40154290199279785,
+      "learning_rate": 7.6086177103270694e-06,
+      "loss": 0.0688913869857788,
+      "mean_token_accuracy": 0.9764530754089356,
+      "num_tokens": 8869555.0,
+      "step": 3900
+    },
+    {
+      "entropy": 0.14314913783222438,
+      "epoch": 8.644030668127053,
+      "grad_norm": 0.6097934246063232,
+      "learning_rate": 6.537527227748551e-06,
+      "loss": 0.07303566455841065,
+      "mean_token_accuracy": 0.9746374058723449,
+      "num_tokens": 8977869.0,
+      "step": 3950
+    },
+    {
+      "entropy": 0.13993739932775498,
+      "epoch": 8.75355969331873,
+      "grad_norm": 0.3800851106643677,
+      "learning_rate": 5.543261163292889e-06,
+      "loss": 0.07268502235412598,
+      "mean_token_accuracy": 0.9749921530485153,
+      "num_tokens": 9085599.0,
+      "step": 4000
+    },
+    {
+      "entropy": 0.13490288957953453,
+      "epoch": 8.863088718510404,
+      "grad_norm": 0.4685533940792084,
+      "learning_rate": 4.627269530856084e-06,
+      "loss": 0.07146972179412842,
+      "mean_token_accuracy": 0.9757514011859894,
+      "num_tokens": 9196726.0,
+      "step": 4050
+    },
+    {
+      "entropy": 0.129932743832469,
+      "epoch": 8.97261774370208,
+      "grad_norm": 0.4819929003715515,
+      "learning_rate": 3.7908881907707993e-06,
+      "loss": 0.06786080360412598,
+      "mean_token_accuracy": 0.9767654579877854,
+      "num_tokens": 9314678.0,
+      "step": 4100
+    },
+    {
+      "epoch": 9.0,
+      "eval_entropy": 0.2552800567613708,
+      "eval_loss": 1.2286313772201538,
+      "eval_mean_token_accuracy": 0.8038436865049695,
+      "eval_num_tokens": 9344250.0,
+      "eval_runtime": 44.7709,
+      "eval_samples_per_second": 22.381,
+      "eval_steps_per_second": 2.814,
+      "step": 4113
+    },
+    {
+      "entropy": 0.13529743457382376,
+      "epoch": 9.08105147864184,
+      "grad_norm": 0.29574504494667053,
+      "learning_rate": 3.035336901619573e-06,
+      "loss": 0.06644320011138916,
+      "mean_token_accuracy": 0.9767602567720894,
+      "num_tokens": 9425344.0,
+      "step": 4150
+    },
+    {
+      "entropy": 0.1262604135274887,
+      "epoch": 9.190580503833516,
+      "grad_norm": 0.357888787984848,
+      "learning_rate": 2.3617175413680483e-06,
+      "loss": 0.06204418659210205,
+      "mean_token_accuracy": 0.9781803792715072,
+      "num_tokens": 9542855.0,
+      "step": 4200
+    },
+    {
+      "entropy": 0.13393638499081134,
+      "epoch": 9.300109529025193,
+      "grad_norm": 0.3549383580684662,
+      "learning_rate": 1.7710125004123868e-06,
+      "loss": 0.06644927978515625,
+      "mean_token_accuracy": 0.9768144589662552,
+      "num_tokens": 9652860.0,
+      "step": 4250
+    },
+    {
+      "entropy": 0.13240817666053772,
+      "epoch": 9.409638554216867,
+      "grad_norm": 0.3490957021713257,
+      "learning_rate": 1.2640832488846195e-06,
+      "loss": 0.06673646450042725,
+      "mean_token_accuracy": 0.976423813700676,
+      "num_tokens": 9764146.0,
+      "step": 4300
+    },
+    {
+      "entropy": 0.12713259197771548,
+      "epoch": 9.519167579408544,
+      "grad_norm": 0.3148532807826996,
+      "learning_rate": 8.416690803051135e-07,
+      "loss": 0.06503040790557861,
+      "mean_token_accuracy": 0.9773415356874466,
+      "num_tokens": 9877849.0,
+      "step": 4350
+    },
+    {
+      "entropy": 0.1263791634887457,
+      "epoch": 9.628696604600218,
+      "grad_norm": 0.32468095421791077,
+      "learning_rate": 5.043860334146096e-07,
+      "loss": 0.06368029594421387,
+      "mean_token_accuracy": 0.9777095991373063,
+      "num_tokens": 9991781.0,
+      "step": 4400
+    },
+    {
+      "entropy": 0.12209340415894986,
+      "epoch": 9.738225629791895,
+      "grad_norm": 0.262802392244339,
+      "learning_rate": 2.527259937579857e-07,
+      "loss": 0.06210928916931152,
+      "mean_token_accuracy": 0.9784713751077652,
+      "num_tokens": 10110574.0,
+      "step": 4450
+    },
+    {
+      "entropy": 0.12839907553046942,
+      "epoch": 9.847754654983571,
+      "grad_norm": 0.23218098282814026,
+      "learning_rate": 8.705597633012421e-08,
+      "loss": 0.06389456272125243,
+      "mean_token_accuracy": 0.9774086111783982,
+      "num_tokens": 10225179.0,
+      "step": 4500
+    },
+    {
+      "entropy": 0.12984012797474861,
+      "epoch": 9.957283680175246,
+      "grad_norm": 0.39813587069511414,
+      "learning_rate": 7.617590330024781e-09,
+      "loss": 0.06630596160888672,
+      "mean_token_accuracy": 0.9772249406576157,
+      "num_tokens": 10336393.0,
+      "step": 4550
+    },
+    {
+      "epoch": 10.0,
+      "eval_entropy": 0.25165843029343893,
+      "eval_loss": 1.2820760011672974,
+      "eval_mean_token_accuracy": 0.8029140503633589,
+      "eval_num_tokens": 10382500.0,
+      "eval_runtime": 44.7381,
+      "eval_samples_per_second": 22.397,
+      "eval_steps_per_second": 2.816,
+      "step": 4570
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 4570,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.2535744725559424e+17,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-914/README.md b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-914/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..7ac50e2706e5b4eae8f28e204601afb241c49e55
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-914/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-2B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-2B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-914/adapter_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-914/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..86d878f780798dd8f06b10d3bdb6eaee702b2da7
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test1/checkpoint-914/adapter_config.json
@@ -0,0 +1,48 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.01686122009693236,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "v_proj",
+    "o_proj",
+    "k_proj",
+    "up_proj",
+    "q_proj",
+    "down_proj",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-3656/trainer_state.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-3656/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..7f38eaa2b90d9297e7da6dc9e14b27ab5a2d5724
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-3656/trainer_state.json
@@ -0,0 +1,852 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 8.0,
+  "eval_steps": 500,
+  "global_step": 3656,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 1.664013209939003,
+      "epoch": 0.10952902519167579,
+      "grad_norm": 1.825055480003357,
+      "learning_rate": 4.4192733942388133e-05,
+      "loss": 1.5422621154785157,
+      "mean_token_accuracy": 0.6574381196498871,
+      "num_tokens": 113087.0,
+      "step": 50
+    },
+    {
+      "entropy": 0.9298514115810395,
+      "epoch": 0.21905805038335158,
+      "grad_norm": 1.4702502489089966,
+      "learning_rate": 8.928736041421276e-05,
+      "loss": 0.8554936218261718,
+      "mean_token_accuracy": 0.7638577824831009,
+      "num_tokens": 230392.0,
+      "step": 100
+    },
+    {
+      "entropy": 0.8588321554660797,
+      "epoch": 0.32858707557502737,
+      "grad_norm": 1.1502039432525635,
+      "learning_rate": 0.00013438198688603737,
+      "loss": 0.7965679168701172,
+      "mean_token_accuracy": 0.7794932717084885,
+      "num_tokens": 341746.0,
+      "step": 150
+    },
+    {
+      "entropy": 0.8281867235898972,
+      "epoch": 0.43811610076670315,
+      "grad_norm": 1.3328826427459717,
+      "learning_rate": 0.00017947661335786198,
+      "loss": 0.7692495727539063,
+      "mean_token_accuracy": 0.7842007935047149,
+      "num_tokens": 460483.0,
+      "step": 200
+    },
+    {
+      "entropy": 0.8131981420516968,
+      "epoch": 0.547645125958379,
+      "grad_norm": 1.2384045124053955,
+      "learning_rate": 0.00022457123982968662,
+      "loss": 0.7567300415039062,
+      "mean_token_accuracy": 0.7894466006755829,
+      "num_tokens": 570627.0,
+      "step": 250
+    },
+    {
+      "entropy": 0.8158841520547867,
+      "epoch": 0.6571741511500547,
+      "grad_norm": 1.29042387008667,
+      "learning_rate": 0.00026966586630151125,
+      "loss": 0.762672119140625,
+      "mean_token_accuracy": 0.7883876228332519,
+      "num_tokens": 677217.0,
+      "step": 300
+    },
+    {
+      "entropy": 0.8154043215513229,
+      "epoch": 0.7667031763417306,
+      "grad_norm": 1.4341716766357422,
+      "learning_rate": 0.00031476049277333584,
+      "loss": 0.7649243927001953,
+      "mean_token_accuracy": 0.7877721953392028,
+      "num_tokens": 790185.0,
+      "step": 350
+    },
+    {
+      "entropy": 0.8017062950134277,
+      "epoch": 0.8762322015334063,
+      "grad_norm": 1.2653132677078247,
+      "learning_rate": 0.0003598551192451605,
+      "loss": 0.7603777313232422,
+      "mean_token_accuracy": 0.7864453828334809,
+      "num_tokens": 910080.0,
+      "step": 400
+    },
+    {
+      "entropy": 0.8266482776403428,
+      "epoch": 0.9857612267250822,
+      "grad_norm": 1.1987895965576172,
+      "learning_rate": 0.0004049497457169851,
+      "loss": 0.783993911743164,
+      "mean_token_accuracy": 0.7832800775766373,
+      "num_tokens": 1023473.0,
+      "step": 450
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.8228033265424153,
+      "eval_loss": 0.8296912908554077,
+      "eval_mean_token_accuracy": 0.769342744634265,
+      "eval_num_tokens": 1038250.0,
+      "eval_runtime": 44.5194,
+      "eval_samples_per_second": 22.485,
+      "eval_steps_per_second": 2.83,
+      "step": 457
+    },
+    {
+      "entropy": 0.8000734815693865,
+      "epoch": 1.0941949616648412,
+      "grad_norm": 1.624888300895691,
+      "learning_rate": 0.00041205884962961745,
+      "loss": 0.7572084045410157,
+      "mean_token_accuracy": 0.7857678087070735,
+      "num_tokens": 1128922.0,
+      "step": 500
+    },
+    {
+      "entropy": 0.7889713329076767,
+      "epoch": 1.203723986856517,
+      "grad_norm": 1.5562400817871094,
+      "learning_rate": 0.0004116562696726163,
+      "loss": 0.757301025390625,
+      "mean_token_accuracy": 0.7859748858213425,
+      "num_tokens": 1245635.0,
+      "step": 550
+    },
+    {
+      "entropy": 0.7770237320661545,
+      "epoch": 1.3132530120481927,
+      "grad_norm": 1.7213724851608276,
+      "learning_rate": 0.00041095388575366294,
+      "loss": 0.7583953857421875,
+      "mean_token_accuracy": 0.7878059667348861,
+      "num_tokens": 1354885.0,
+      "step": 600
+    },
+    {
+      "entropy": 0.7724025696516037,
+      "epoch": 1.4227820372398685,
+      "grad_norm": 1.2281577587127686,
+      "learning_rate": 0.00040995272221269925,
+      "loss": 0.7478806304931641,
+      "mean_token_accuracy": 0.789793593287468,
+      "num_tokens": 1472232.0,
+      "step": 650
+    },
+    {
+      "entropy": 0.7704478764533996,
+      "epoch": 1.5323110624315444,
+      "grad_norm": 1.5951343774795532,
+      "learning_rate": 0.00040865423912273625,
+      "loss": 0.7493467712402344,
+      "mean_token_accuracy": 0.7901222252845764,
+      "num_tokens": 1587494.0,
+      "step": 700
+    },
+    {
+      "entropy": 0.7755234348773956,
+      "epoch": 1.6418400876232202,
+      "grad_norm": 1.2122973203659058,
+      "learning_rate": 0.0004070603301605186,
+      "loss": 0.7549277496337891,
+      "mean_token_accuracy": 0.7899289625883102,
+      "num_tokens": 1696927.0,
+      "step": 750
+    },
+    {
+      "entropy": 0.7561545193195343,
+      "epoch": 1.751369112814896,
+      "grad_norm": 1.262399673461914,
+      "learning_rate": 0.00040517331984483133,
+      "loss": 0.7413782501220703,
+      "mean_token_accuracy": 0.7917980045080185,
+      "num_tokens": 1813415.0,
+      "step": 800
+    },
+    {
+      "entropy": 0.7466552203893662,
+      "epoch": 1.8608981380065717,
+      "grad_norm": 1.2854546308517456,
+      "learning_rate": 0.00040299596014647715,
+      "loss": 0.7443357849121094,
+      "mean_token_accuracy": 0.79294473528862,
+      "num_tokens": 1924911.0,
+      "step": 850
+    },
+    {
+      "entropy": 0.7460890018939972,
+      "epoch": 1.9704271631982475,
+      "grad_norm": 1.0535873174667358,
+      "learning_rate": 0.00040053142647486815,
+      "loss": 0.742437973022461,
+      "mean_token_accuracy": 0.7911104756593704,
+      "num_tokens": 2044650.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.0,
+      "eval_entropy": 0.7546336958332668,
+      "eval_loss": 0.7908218502998352,
+      "eval_mean_token_accuracy": 0.7816079741432553,
+      "eval_num_tokens": 2076500.0,
+      "eval_runtime": 43.6577,
+      "eval_samples_per_second": 22.928,
+      "eval_steps_per_second": 2.886,
+      "step": 914
+    },
+    {
+      "entropy": 0.6920583218035071,
+      "epoch": 2.078860898138007,
+      "grad_norm": 0.8588879704475403,
+      "learning_rate": 0.0003977833130470843,
+      "loss": 0.6903114318847656,
+      "mean_token_accuracy": 0.7997718549738027,
+      "num_tokens": 2160498.0,
+      "step": 950
+    },
+    {
+      "entropy": 0.6784634017944335,
+      "epoch": 2.1883899233296824,
+      "grad_norm": 1.1033616065979004,
+      "learning_rate": 0.0003947556276461532,
+      "loss": 0.6780696868896484,
+      "mean_token_accuracy": 0.8036011070013046,
+      "num_tokens": 2269979.0,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6898973470926285,
+      "epoch": 2.297918948521358,
+      "grad_norm": 1.1214436292648315,
+      "learning_rate": 0.0003914527857761949,
+      "loss": 0.6930846405029297,
+      "mean_token_accuracy": 0.8000095456838607,
+      "num_tokens": 2378362.0,
+      "step": 1050
+    },
+    {
+      "entropy": 0.6754742419719696,
+      "epoch": 2.407447973713034,
+      "grad_norm": 1.0675715208053589,
+      "learning_rate": 0.0003878796042229559,
+      "loss": 0.6745318603515625,
+      "mean_token_accuracy": 0.8050142484903335,
+      "num_tokens": 2498494.0,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6677694636583328,
+      "epoch": 2.5169769989047097,
+      "grad_norm": 1.7671644687652588,
+      "learning_rate": 0.00038404129402912424,
+      "loss": 0.6703590393066406,
+      "mean_token_accuracy": 0.8045552480220795,
+      "num_tokens": 2611646.0,
+      "step": 1150
+    },
+    {
+      "entropy": 0.6750778061151504,
+      "epoch": 2.6265060240963853,
+      "grad_norm": 1.428247332572937,
+      "learning_rate": 0.00037994345289466886,
+      "loss": 0.6790786743164062,
+      "mean_token_accuracy": 0.8028086322546005,
+      "num_tokens": 2717744.0,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6834373778104782,
+      "epoch": 2.7360350492880614,
+      "grad_norm": 0.9393771886825562,
+      "learning_rate": 0.0003755920570132876,
+      "loss": 0.6844132232666016,
+      "mean_token_accuracy": 0.8027357232570648,
+      "num_tokens": 2827750.0,
+      "step": 1250
+    },
+    {
+      "entropy": 0.671256046295166,
+      "epoch": 2.845564074479737,
+      "grad_norm": 1.2709442377090454,
+      "learning_rate": 0.0003709934523568692,
+      "loss": 0.6658607482910156,
+      "mean_token_accuracy": 0.8065212601423264,
+      "num_tokens": 2948958.0,
+      "step": 1300
+    },
+    {
+      "entropy": 0.6620459565520287,
+      "epoch": 2.955093099671413,
+      "grad_norm": 1.0260528326034546,
+      "learning_rate": 0.00036615434542067925,
+      "loss": 0.6596417236328125,
+      "mean_token_accuracy": 0.8089053362607956,
+      "num_tokens": 3066917.0,
+      "step": 1350
+    },
+    {
+      "epoch": 3.0,
+      "eval_entropy": 0.6772078592625875,
+      "eval_loss": 0.7904035449028015,
+      "eval_mean_token_accuracy": 0.7809106636615026,
+      "eval_num_tokens": 3114750.0,
+      "eval_runtime": 44.0158,
+      "eval_samples_per_second": 22.742,
+      "eval_steps_per_second": 2.863,
+      "step": 1371
+    },
+    {
+      "entropy": 0.6129003967901673,
+      "epoch": 3.063526834611172,
+      "grad_norm": 1.0416573286056519,
+      "learning_rate": 0.00036108179344276724,
+      "loss": 0.6105740356445313,
+      "mean_token_accuracy": 0.8168958028157552,
+      "num_tokens": 3180366.0,
+      "step": 1400
+    },
+    {
+      "entropy": 0.6044845461845398,
+      "epoch": 3.1730558598028478,
+      "grad_norm": 0.9305349588394165,
+      "learning_rate": 0.00035578319411185954,
+      "loss": 0.59905029296875,
+      "mean_token_accuracy": 0.8190373265743256,
+      "num_tokens": 3293918.0,
+      "step": 1450
+    },
+    {
+      "entropy": 0.6040669977664948,
+      "epoch": 3.2825848849945234,
+      "grad_norm": 1.039084553718567,
+      "learning_rate": 0.00035026627477874664,
+      "loss": 0.5956837463378907,
+      "mean_token_accuracy": 0.8208023190498352,
+      "num_tokens": 3409058.0,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5916981258988381,
+      "epoch": 3.3921139101861995,
+      "grad_norm": 1.1696960926055908,
+      "learning_rate": 0.00034453908118689904,
+      "loss": 0.5919752502441407,
+      "mean_token_accuracy": 0.8211179333925247,
+      "num_tokens": 3525131.0,
+      "step": 1550
+    },
+    {
+      "entropy": 0.6082543370127678,
+      "epoch": 3.501642935377875,
+      "grad_norm": 1.0947457551956177,
+      "learning_rate": 0.0003386099657387468,
+      "loss": 0.6049380493164063,
+      "mean_token_accuracy": 0.818329308629036,
+      "num_tokens": 3641504.0,
+      "step": 1600
+    },
+    {
+      "entropy": 0.6244189664721489,
+      "epoch": 3.6111719605695507,
+      "grad_norm": 1.2528455257415771,
+      "learning_rate": 0.00033248757531473597,
+      "loss": 0.6117112731933594,
+      "mean_token_accuracy": 0.8165279686450958,
+      "num_tokens": 3752182.0,
+      "step": 1650
+    },
+    {
+      "entropy": 0.6266464674472809,
+      "epoch": 3.7207009857612268,
+      "grad_norm": 0.8912960290908813,
+      "learning_rate": 0.00032618083866292364,
+      "loss": 0.6233459854125977,
+      "mean_token_accuracy": 0.8161942487955094,
+      "num_tokens": 3866473.0,
+      "step": 1700
+    },
+    {
+      "entropy": 0.603082166314125,
+      "epoch": 3.8302300109529024,
+      "grad_norm": 0.9780993461608887,
+      "learning_rate": 0.00031969895337750556,
+      "loss": 0.5993933868408203,
+      "mean_token_accuracy": 0.8191226273775101,
+      "num_tokens": 3977393.0,
+      "step": 1750
+    },
+    {
+      "entropy": 0.6284996509552002,
+      "epoch": 3.9397590361445785,
+      "grad_norm": 1.3573766946792603,
+      "learning_rate": 0.00031305137248526394,
+      "loss": 0.6139850616455078,
+      "mean_token_accuracy": 0.8174162280559539,
+      "num_tokens": 4088936.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5731133787877976,
+      "eval_loss": 0.7714413404464722,
+      "eval_mean_token_accuracy": 0.7904208685670581,
+      "eval_num_tokens": 4153000.0,
+      "eval_runtime": 43.8857,
+      "eval_samples_per_second": 22.809,
+      "eval_steps_per_second": 2.871,
+      "step": 1828
+    },
+    {
+      "entropy": 0.56682777645612,
+      "epoch": 4.048192771084337,
+      "grad_norm": 0.9725717306137085,
+      "learning_rate": 0.0003062477906594989,
+      "loss": 0.5487574005126953,
+      "mean_token_accuracy": 0.8309493137128425,
+      "num_tokens": 4202329.0,
+      "step": 1850
+    },
+    {
+      "entropy": 0.522312518954277,
+      "epoch": 4.157721796276014,
+      "grad_norm": 1.0364148616790771,
+      "learning_rate": 0.00029929813008154865,
+      "loss": 0.5065541458129883,
+      "mean_token_accuracy": 0.8395609927177429,
+      "num_tokens": 4320321.0,
+      "step": 1900
+    },
+    {
+      "entropy": 0.5235522103309631,
+      "epoch": 4.267250821467689,
+      "grad_norm": 0.9119980931282043,
+      "learning_rate": 0.00029221252597051635,
+      "loss": 0.5116676712036132,
+      "mean_token_accuracy": 0.8383910119533539,
+      "num_tokens": 4435139.0,
+      "step": 1950
+    },
+    {
+      "entropy": 0.5248301169276237,
+      "epoch": 4.376779846659365,
+      "grad_norm": 1.2733819484710693,
+      "learning_rate": 0.00028500131180230935,
+      "loss": 0.5151065826416016,
+      "mean_token_accuracy": 0.8378184515237809,
+      "num_tokens": 4547272.0,
+      "step": 2000
+    },
+    {
+      "entropy": 0.531632961332798,
+      "epoch": 4.48630887185104,
+      "grad_norm": 1.008576512336731,
+      "learning_rate": 0.00027767500423954395,
+      "loss": 0.5181634902954102,
+      "mean_token_accuracy": 0.8388127386569977,
+      "num_tokens": 4659036.0,
+      "step": 2050
+    },
+    {
+      "entropy": 0.5213253930211067,
+      "epoch": 4.595837897042716,
+      "grad_norm": 0.9017001986503601,
+      "learning_rate": 0.0002702442877942959,
+      "loss": 0.5104235458374023,
+      "mean_token_accuracy": 0.8390659308433532,
+      "num_tokens": 4769957.0,
+      "step": 2100
+    },
+    {
+      "entropy": 0.5312498080730438,
+      "epoch": 4.705366922234392,
+      "grad_norm": 0.8270431756973267,
+      "learning_rate": 0.00026271999924606297,
+      "loss": 0.5256425476074219,
+      "mean_token_accuracy": 0.8363041788339615,
+      "num_tokens": 4882487.0,
+      "step": 2150
+    },
+    {
+      "entropy": 0.5297493544220925,
+      "epoch": 4.814895947426068,
+      "grad_norm": 0.6612098813056946,
+      "learning_rate": 0.00025511311183766457,
+      "loss": 0.5210869979858398,
+      "mean_token_accuracy": 0.837263793349266,
+      "num_tokens": 4993060.0,
+      "step": 2200
+    },
+    {
+      "entropy": 0.5250286677479744,
+      "epoch": 4.924424972617744,
+      "grad_norm": 0.8537449240684509,
+      "learning_rate": 0.00024743471927212595,
+      "loss": 0.5147652816772461,
+      "mean_token_accuracy": 0.8392492067813874,
+      "num_tokens": 5111182.0,
+      "step": 2250
+    },
+    {
+      "epoch": 5.0,
+      "eval_entropy": 0.5230902518544879,
+      "eval_loss": 0.7304916977882385,
+      "eval_mean_token_accuracy": 0.8025014537667471,
+      "eval_num_tokens": 5191250.0,
+      "eval_runtime": 43.9445,
+      "eval_samples_per_second": 22.779,
+      "eval_steps_per_second": 2.867,
+      "step": 2285
+    },
+    {
+      "entropy": 0.48566745266769873,
+      "epoch": 5.032858707557502,
+      "grad_norm": 0.7822101712226868,
+      "learning_rate": 0.000239696019533887,
+      "loss": 0.4735472869873047,
+      "mean_token_accuracy": 0.8504744623646592,
+      "num_tokens": 5225197.0,
+      "step": 2300
+    },
+    {
+      "entropy": 0.40831263691186903,
+      "epoch": 5.142387732749179,
+      "grad_norm": 0.8123568296432495,
+      "learning_rate": 0.00023190829855792867,
+      "loss": 0.3874293518066406,
+      "mean_token_accuracy": 0.8715608316659927,
+      "num_tokens": 5336728.0,
+      "step": 2350
+    },
+    {
+      "entropy": 0.41975140184164045,
+      "epoch": 5.2519167579408546,
+      "grad_norm": 0.7969071865081787,
+      "learning_rate": 0.00022408291377063483,
+      "loss": 0.40011211395263674,
+      "mean_token_accuracy": 0.8658375990390778,
+      "num_tokens": 5446986.0,
+      "step": 2400
+    },
+    {
+      "entropy": 0.41357740610837934,
+      "epoch": 5.36144578313253,
+      "grad_norm": 1.0098742246627808,
+      "learning_rate": 0.000216231277526393,
+      "loss": 0.39797264099121094,
+      "mean_token_accuracy": 0.8674204486608506,
+      "num_tokens": 5563479.0,
+      "step": 2450
+    },
+    {
+      "entropy": 0.4265018665790558,
+      "epoch": 5.470974808324206,
+      "grad_norm": 1.160515308380127,
+      "learning_rate": 0.00020836484046408868,
+      "loss": 0.40515335083007814,
+      "mean_token_accuracy": 0.8648038375377655,
+      "num_tokens": 5677338.0,
+      "step": 2500
+    },
+    {
+      "entropy": 0.4268675494194031,
+      "epoch": 5.580503833515881,
+      "grad_norm": 0.7267200946807861,
+      "learning_rate": 0.0002004950748077673,
+      "loss": 0.40761962890625,
+      "mean_token_accuracy": 0.864390144944191,
+      "num_tokens": 5787988.0,
+      "step": 2550
+    },
+    {
+      "entropy": 0.4280226981639862,
+      "epoch": 5.690032858707557,
+      "grad_norm": 0.8958321809768677,
+      "learning_rate": 0.00019263345763581628,
+      "loss": 0.40727943420410156,
+      "mean_token_accuracy": 0.8640619122982025,
+      "num_tokens": 5902559.0,
+      "step": 2600
+    },
+    {
+      "entropy": 0.431531577706337,
+      "epoch": 5.7995618838992335,
+      "grad_norm": 0.875791072845459,
+      "learning_rate": 0.00018479145414306803,
+      "loss": 0.40996681213378905,
+      "mean_token_accuracy": 0.8628572630882263,
+      "num_tokens": 6014039.0,
+      "step": 2650
+    },
+    {
+      "entropy": 0.41479706585407256,
+      "epoch": 5.909090909090909,
+      "grad_norm": 0.7954572439193726,
+      "learning_rate": 0.0001769805009202332,
+      "loss": 0.3963738250732422,
+      "mean_token_accuracy": 0.8696580404043197,
+      "num_tokens": 6134464.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.0,
+      "eval_entropy": 0.472232200796642,
+      "eval_loss": 0.747288703918457,
+      "eval_mean_token_accuracy": 0.802532631726492,
+      "eval_num_tokens": 6229500.0,
+      "eval_runtime": 43.8095,
+      "eval_samples_per_second": 22.849,
+      "eval_steps_per_second": 2.876,
+      "step": 2742
+    },
+    {
+      "entropy": 0.405825153143719,
+      "epoch": 6.017524644030668,
+      "grad_norm": 0.675167977809906,
+      "learning_rate": 0.00016921198927505012,
+      "loss": 0.38604618072509767,
+      "mean_token_accuracy": 0.8706694215235083,
+      "num_tokens": 6249905.0,
+      "step": 2750
+    },
+    {
+      "entropy": 0.28874062046408655,
+      "epoch": 6.127053669222344,
+      "grad_norm": 0.8683274984359741,
+      "learning_rate": 0.0001614972486194732,
+      "loss": 0.2581937789916992,
+      "mean_token_accuracy": 0.9093546283245086,
+      "num_tokens": 6367665.0,
+      "step": 2800
+    },
+    {
+      "entropy": 0.29525566384196283,
+      "epoch": 6.23658269441402,
+      "grad_norm": 0.8548017740249634,
+      "learning_rate": 0.0001538475299471291,
+      "loss": 0.26408302307128906,
+      "mean_token_accuracy": 0.9077397453784942,
+      "num_tokens": 6481769.0,
+      "step": 2850
+    },
+    {
+      "entropy": 0.30853822842240336,
+      "epoch": 6.3461117196056955,
+      "grad_norm": 0.9827961325645447,
+      "learning_rate": 0.00014627398942513635,
+      "loss": 0.2714972877502441,
+      "mean_token_accuracy": 0.9044403052330017,
+      "num_tokens": 6595712.0,
+      "step": 2900
+    },
+    {
+      "entropy": 0.30256781667470933,
+      "epoch": 6.455640744797371,
+      "grad_norm": 0.8727903962135315,
+      "learning_rate": 0.0001387876721242171,
+      "loss": 0.2691990852355957,
+      "mean_token_accuracy": 0.9049568754434586,
+      "num_tokens": 6708841.0,
+      "step": 2950
+    },
+    {
+      "entropy": 0.305324949324131,
+      "epoch": 6.565169769989047,
+      "grad_norm": 0.7359956502914429,
+      "learning_rate": 0.00013139949591082985,
+      "loss": 0.274547176361084,
+      "mean_token_accuracy": 0.9040277010202408,
+      "num_tokens": 6818025.0,
+      "step": 3000
+    },
+    {
+      "entropy": 0.3107963638007641,
+      "epoch": 6.674698795180722,
+      "grad_norm": 0.814873456954956,
+      "learning_rate": 0.0001241202355248137,
+      "loss": 0.27763776779174804,
+      "mean_token_accuracy": 0.9029175835847855,
+      "num_tokens": 6929746.0,
+      "step": 3050
+    },
+    {
+      "entropy": 0.30067121908068656,
+      "epoch": 6.784227820372399,
+      "grad_norm": 0.796090841293335,
+      "learning_rate": 0.00011696050686576498,
+      "loss": 0.26881502151489256,
+      "mean_token_accuracy": 0.9052424907684327,
+      "num_tokens": 7046109.0,
+      "step": 3100
+    },
+    {
+      "entropy": 0.30515122786164284,
+      "epoch": 6.8937568455640745,
+      "grad_norm": 0.8339642286300659,
+      "learning_rate": 0.00010993075151106208,
+      "loss": 0.273194522857666,
+      "mean_token_accuracy": 0.904237716794014,
+      "num_tokens": 7156155.0,
+      "step": 3150
+    },
+    {
+      "epoch": 7.0,
+      "eval_entropy": 0.37887449586202226,
+      "eval_loss": 0.8343665599822998,
+      "eval_mean_token_accuracy": 0.8046237868922097,
+      "eval_num_tokens": 7267750.0,
+      "eval_runtime": 43.2059,
+      "eval_samples_per_second": 23.168,
+      "eval_steps_per_second": 2.916,
+      "step": 3199
+    },
+    {
+      "entropy": 0.29947714342011345,
+      "epoch": 7.002190580503833,
+      "grad_norm": 0.5025448799133301,
+      "learning_rate": 0.0001030412214881193,
+      "loss": 0.2715941619873047,
+      "mean_token_accuracy": 0.9054891123916163,
+      "num_tokens": 7269424.0,
+      "step": 3200
+    },
+    {
+      "entropy": 0.19013839438557625,
+      "epoch": 7.11171960569551,
+      "grad_norm": 0.7953240275382996,
+      "learning_rate": 9.630196432307369e-05,
+      "loss": 0.15198980331420897,
+      "mean_token_accuracy": 0.9485088515281678,
+      "num_tokens": 7382588.0,
+      "step": 3250
+    },
+    {
+      "entropy": 0.18270175024867058,
+      "epoch": 7.221248630887185,
+      "grad_norm": 0.8207944631576538,
+      "learning_rate": 8.972280838771363e-05,
+      "loss": 0.1504285717010498,
+      "mean_token_accuracy": 0.9483412396907807,
+      "num_tokens": 7494048.0,
+      "step": 3300
+    },
+    {
+      "entropy": 0.1822182248532772,
+      "epoch": 7.330777656078861,
+      "grad_norm": 0.7845531105995178,
+      "learning_rate": 8.331334856601539e-05,
+      "loss": 0.1527143096923828,
+      "mean_token_accuracy": 0.9482884389162064,
+      "num_tokens": 7607343.0,
+      "step": 3350
+    },
+    {
+      "entropy": 0.18213839918375016,
+      "epoch": 7.4403066812705365,
+      "grad_norm": 0.9096264839172363,
+      "learning_rate": 7.708293226119355e-05,
+      "loss": 0.1504144287109375,
+      "mean_token_accuracy": 0.9482587432861328,
+      "num_tokens": 7724520.0,
+      "step": 3400
+    },
+    {
+      "entropy": 0.18474044188857078,
+      "epoch": 7.549835706462212,
+      "grad_norm": 0.7610820531845093,
+      "learning_rate": 7.104064576367165e-05,
+      "loss": 0.15584409713745118,
+      "mean_token_accuracy": 0.9465331947803497,
+      "num_tokens": 7836568.0,
+      "step": 3450
+    },
+    {
+      "entropy": 0.1816471180319786,
+      "epoch": 7.659364731653889,
+      "grad_norm": 0.8613850474357605,
+      "learning_rate": 6.519530099985234e-05,
+      "loss": 0.15603148460388183,
+      "mean_token_accuracy": 0.9455479162931443,
+      "num_tokens": 7947988.0,
+      "step": 3500
+    },
+    {
+      "entropy": 0.18248806044459343,
+      "epoch": 7.768893756845564,
+      "grad_norm": 0.644688069820404,
+      "learning_rate": 5.9555422681015177e-05,
+      "loss": 0.1541291046142578,
+      "mean_token_accuracy": 0.9463654214143753,
+      "num_tokens": 8066809.0,
+      "step": 3550
+    },
+    {
+      "entropy": 0.1824373207241297,
+      "epoch": 7.87842278203724,
+      "grad_norm": 0.7455086708068848,
+      "learning_rate": 5.4129235871080286e-05,
+      "loss": 0.15226722717285157,
+      "mean_token_accuracy": 0.9470921188592911,
+      "num_tokens": 8178429.0,
+      "step": 3600
+    },
+    {
+      "entropy": 0.17558989696204663,
+      "epoch": 7.9879518072289155,
+      "grad_norm": 0.6741284728050232,
+      "learning_rate": 4.8924653991372106e-05,
+      "loss": 0.14881887435913085,
+      "mean_token_accuracy": 0.9479702442884446,
+      "num_tokens": 8294498.0,
+      "step": 3650
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.31352401418345316,
+      "eval_loss": 0.9514623284339905,
+      "eval_mean_token_accuracy": 0.8020065157186418,
+      "eval_num_tokens": 8306000.0,
+      "eval_runtime": 43.9551,
+      "eval_samples_per_second": 22.773,
+      "eval_steps_per_second": 2.867,
+      "step": 3656
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 4570,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.8007844674315584e+17,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4113/README.md b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4113/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..7ac50e2706e5b4eae8f28e204601afb241c49e55
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4113/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-2B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-2B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4113/adapter_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4113/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..4d005ec8c649e4abb0922fac729d1444177e0dc1
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4113/adapter_config.json
@@ -0,0 +1,48 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.0029143607590526367,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "v_proj",
+    "o_proj",
+    "k_proj",
+    "up_proj",
+    "q_proj",
+    "down_proj",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4113/chat_template.jinja b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4113/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..0ef09f214eaa6d9bca297988afc1454b5827b2c7
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4113/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is true %}
+        {{- '<think>\n' }}
+    {%- else %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4113/tokenizer_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4113/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4113/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4113/trainer_state.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4113/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..2354ed980c09515cebf9c5c73e16aadca2fcada3
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4113/trainer_state.json
@@ -0,0 +1,953 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 9.0,
+  "eval_steps": 500,
+  "global_step": 4113,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 1.664013209939003,
+      "epoch": 0.10952902519167579,
+      "grad_norm": 1.825055480003357,
+      "learning_rate": 4.4192733942388133e-05,
+      "loss": 1.5422621154785157,
+      "mean_token_accuracy": 0.6574381196498871,
+      "num_tokens": 113087.0,
+      "step": 50
+    },
+    {
+      "entropy": 0.9298514115810395,
+      "epoch": 0.21905805038335158,
+      "grad_norm": 1.4702502489089966,
+      "learning_rate": 8.928736041421276e-05,
+      "loss": 0.8554936218261718,
+      "mean_token_accuracy": 0.7638577824831009,
+      "num_tokens": 230392.0,
+      "step": 100
+    },
+    {
+      "entropy": 0.8588321554660797,
+      "epoch": 0.32858707557502737,
+      "grad_norm": 1.1502039432525635,
+      "learning_rate": 0.00013438198688603737,
+      "loss": 0.7965679168701172,
+      "mean_token_accuracy": 0.7794932717084885,
+      "num_tokens": 341746.0,
+      "step": 150
+    },
+    {
+      "entropy": 0.8281867235898972,
+      "epoch": 0.43811610076670315,
+      "grad_norm": 1.3328826427459717,
+      "learning_rate": 0.00017947661335786198,
+      "loss": 0.7692495727539063,
+      "mean_token_accuracy": 0.7842007935047149,
+      "num_tokens": 460483.0,
+      "step": 200
+    },
+    {
+      "entropy": 0.8131981420516968,
+      "epoch": 0.547645125958379,
+      "grad_norm": 1.2384045124053955,
+      "learning_rate": 0.00022457123982968662,
+      "loss": 0.7567300415039062,
+      "mean_token_accuracy": 0.7894466006755829,
+      "num_tokens": 570627.0,
+      "step": 250
+    },
+    {
+      "entropy": 0.8158841520547867,
+      "epoch": 0.6571741511500547,
+      "grad_norm": 1.29042387008667,
+      "learning_rate": 0.00026966586630151125,
+      "loss": 0.762672119140625,
+      "mean_token_accuracy": 0.7883876228332519,
+      "num_tokens": 677217.0,
+      "step": 300
+    },
+    {
+      "entropy": 0.8154043215513229,
+      "epoch": 0.7667031763417306,
+      "grad_norm": 1.4341716766357422,
+      "learning_rate": 0.00031476049277333584,
+      "loss": 0.7649243927001953,
+      "mean_token_accuracy": 0.7877721953392028,
+      "num_tokens": 790185.0,
+      "step": 350
+    },
+    {
+      "entropy": 0.8017062950134277,
+      "epoch": 0.8762322015334063,
+      "grad_norm": 1.2653132677078247,
+      "learning_rate": 0.0003598551192451605,
+      "loss": 0.7603777313232422,
+      "mean_token_accuracy": 0.7864453828334809,
+      "num_tokens": 910080.0,
+      "step": 400
+    },
+    {
+      "entropy": 0.8266482776403428,
+      "epoch": 0.9857612267250822,
+      "grad_norm": 1.1987895965576172,
+      "learning_rate": 0.0004049497457169851,
+      "loss": 0.783993911743164,
+      "mean_token_accuracy": 0.7832800775766373,
+      "num_tokens": 1023473.0,
+      "step": 450
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.8228033265424153,
+      "eval_loss": 0.8296912908554077,
+      "eval_mean_token_accuracy": 0.769342744634265,
+      "eval_num_tokens": 1038250.0,
+      "eval_runtime": 44.5194,
+      "eval_samples_per_second": 22.485,
+      "eval_steps_per_second": 2.83,
+      "step": 457
+    },
+    {
+      "entropy": 0.8000734815693865,
+      "epoch": 1.0941949616648412,
+      "grad_norm": 1.624888300895691,
+      "learning_rate": 0.00041205884962961745,
+      "loss": 0.7572084045410157,
+      "mean_token_accuracy": 0.7857678087070735,
+      "num_tokens": 1128922.0,
+      "step": 500
+    },
+    {
+      "entropy": 0.7889713329076767,
+      "epoch": 1.203723986856517,
+      "grad_norm": 1.5562400817871094,
+      "learning_rate": 0.0004116562696726163,
+      "loss": 0.757301025390625,
+      "mean_token_accuracy": 0.7859748858213425,
+      "num_tokens": 1245635.0,
+      "step": 550
+    },
+    {
+      "entropy": 0.7770237320661545,
+      "epoch": 1.3132530120481927,
+      "grad_norm": 1.7213724851608276,
+      "learning_rate": 0.00041095388575366294,
+      "loss": 0.7583953857421875,
+      "mean_token_accuracy": 0.7878059667348861,
+      "num_tokens": 1354885.0,
+      "step": 600
+    },
+    {
+      "entropy": 0.7724025696516037,
+      "epoch": 1.4227820372398685,
+      "grad_norm": 1.2281577587127686,
+      "learning_rate": 0.00040995272221269925,
+      "loss": 0.7478806304931641,
+      "mean_token_accuracy": 0.789793593287468,
+      "num_tokens": 1472232.0,
+      "step": 650
+    },
+    {
+      "entropy": 0.7704478764533996,
+      "epoch": 1.5323110624315444,
+      "grad_norm": 1.5951343774795532,
+      "learning_rate": 0.00040865423912273625,
+      "loss": 0.7493467712402344,
+      "mean_token_accuracy": 0.7901222252845764,
+      "num_tokens": 1587494.0,
+      "step": 700
+    },
+    {
+      "entropy": 0.7755234348773956,
+      "epoch": 1.6418400876232202,
+      "grad_norm": 1.2122973203659058,
+      "learning_rate": 0.0004070603301605186,
+      "loss": 0.7549277496337891,
+      "mean_token_accuracy": 0.7899289625883102,
+      "num_tokens": 1696927.0,
+      "step": 750
+    },
+    {
+      "entropy": 0.7561545193195343,
+      "epoch": 1.751369112814896,
+      "grad_norm": 1.262399673461914,
+      "learning_rate": 0.00040517331984483133,
+      "loss": 0.7413782501220703,
+      "mean_token_accuracy": 0.7917980045080185,
+      "num_tokens": 1813415.0,
+      "step": 800
+    },
+    {
+      "entropy": 0.7466552203893662,
+      "epoch": 1.8608981380065717,
+      "grad_norm": 1.2854546308517456,
+      "learning_rate": 0.00040299596014647715,
+      "loss": 0.7443357849121094,
+      "mean_token_accuracy": 0.79294473528862,
+      "num_tokens": 1924911.0,
+      "step": 850
+    },
+    {
+      "entropy": 0.7460890018939972,
+      "epoch": 1.9704271631982475,
+      "grad_norm": 1.0535873174667358,
+      "learning_rate": 0.00040053142647486815,
+      "loss": 0.742437973022461,
+      "mean_token_accuracy": 0.7911104756593704,
+      "num_tokens": 2044650.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.0,
+      "eval_entropy": 0.7546336958332668,
+      "eval_loss": 0.7908218502998352,
+      "eval_mean_token_accuracy": 0.7816079741432553,
+      "eval_num_tokens": 2076500.0,
+      "eval_runtime": 43.6577,
+      "eval_samples_per_second": 22.928,
+      "eval_steps_per_second": 2.886,
+      "step": 914
+    },
+    {
+      "entropy": 0.6920583218035071,
+      "epoch": 2.078860898138007,
+      "grad_norm": 0.8588879704475403,
+      "learning_rate": 0.0003977833130470843,
+      "loss": 0.6903114318847656,
+      "mean_token_accuracy": 0.7997718549738027,
+      "num_tokens": 2160498.0,
+      "step": 950
+    },
+    {
+      "entropy": 0.6784634017944335,
+      "epoch": 2.1883899233296824,
+      "grad_norm": 1.1033616065979004,
+      "learning_rate": 0.0003947556276461532,
+      "loss": 0.6780696868896484,
+      "mean_token_accuracy": 0.8036011070013046,
+      "num_tokens": 2269979.0,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6898973470926285,
+      "epoch": 2.297918948521358,
+      "grad_norm": 1.1214436292648315,
+      "learning_rate": 0.0003914527857761949,
+      "loss": 0.6930846405029297,
+      "mean_token_accuracy": 0.8000095456838607,
+      "num_tokens": 2378362.0,
+      "step": 1050
+    },
+    {
+      "entropy": 0.6754742419719696,
+      "epoch": 2.407447973713034,
+      "grad_norm": 1.0675715208053589,
+      "learning_rate": 0.0003878796042229559,
+      "loss": 0.6745318603515625,
+      "mean_token_accuracy": 0.8050142484903335,
+      "num_tokens": 2498494.0,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6677694636583328,
+      "epoch": 2.5169769989047097,
+      "grad_norm": 1.7671644687652588,
+      "learning_rate": 0.00038404129402912424,
+      "loss": 0.6703590393066406,
+      "mean_token_accuracy": 0.8045552480220795,
+      "num_tokens": 2611646.0,
+      "step": 1150
+    },
+    {
+      "entropy": 0.6750778061151504,
+      "epoch": 2.6265060240963853,
+      "grad_norm": 1.428247332572937,
+      "learning_rate": 0.00037994345289466886,
+      "loss": 0.6790786743164062,
+      "mean_token_accuracy": 0.8028086322546005,
+      "num_tokens": 2717744.0,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6834373778104782,
+      "epoch": 2.7360350492880614,
+      "grad_norm": 0.9393771886825562,
+      "learning_rate": 0.0003755920570132876,
+      "loss": 0.6844132232666016,
+      "mean_token_accuracy": 0.8027357232570648,
+      "num_tokens": 2827750.0,
+      "step": 1250
+    },
+    {
+      "entropy": 0.671256046295166,
+      "epoch": 2.845564074479737,
+      "grad_norm": 1.2709442377090454,
+      "learning_rate": 0.0003709934523568692,
+      "loss": 0.6658607482910156,
+      "mean_token_accuracy": 0.8065212601423264,
+      "num_tokens": 2948958.0,
+      "step": 1300
+    },
+    {
+      "entropy": 0.6620459565520287,
+      "epoch": 2.955093099671413,
+      "grad_norm": 1.0260528326034546,
+      "learning_rate": 0.00036615434542067925,
+      "loss": 0.6596417236328125,
+      "mean_token_accuracy": 0.8089053362607956,
+      "num_tokens": 3066917.0,
+      "step": 1350
+    },
+    {
+      "epoch": 3.0,
+      "eval_entropy": 0.6772078592625875,
+      "eval_loss": 0.7904035449028015,
+      "eval_mean_token_accuracy": 0.7809106636615026,
+      "eval_num_tokens": 3114750.0,
+      "eval_runtime": 44.0158,
+      "eval_samples_per_second": 22.742,
+      "eval_steps_per_second": 2.863,
+      "step": 1371
+    },
+    {
+      "entropy": 0.6129003967901673,
+      "epoch": 3.063526834611172,
+      "grad_norm": 1.0416573286056519,
+      "learning_rate": 0.00036108179344276724,
+      "loss": 0.6105740356445313,
+      "mean_token_accuracy": 0.8168958028157552,
+      "num_tokens": 3180366.0,
+      "step": 1400
+    },
+    {
+      "entropy": 0.6044845461845398,
+      "epoch": 3.1730558598028478,
+      "grad_norm": 0.9305349588394165,
+      "learning_rate": 0.00035578319411185954,
+      "loss": 0.59905029296875,
+      "mean_token_accuracy": 0.8190373265743256,
+      "num_tokens": 3293918.0,
+      "step": 1450
+    },
+    {
+      "entropy": 0.6040669977664948,
+      "epoch": 3.2825848849945234,
+      "grad_norm": 1.039084553718567,
+      "learning_rate": 0.00035026627477874664,
+      "loss": 0.5956837463378907,
+      "mean_token_accuracy": 0.8208023190498352,
+      "num_tokens": 3409058.0,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5916981258988381,
+      "epoch": 3.3921139101861995,
+      "grad_norm": 1.1696960926055908,
+      "learning_rate": 0.00034453908118689904,
+      "loss": 0.5919752502441407,
+      "mean_token_accuracy": 0.8211179333925247,
+      "num_tokens": 3525131.0,
+      "step": 1550
+    },
+    {
+      "entropy": 0.6082543370127678,
+      "epoch": 3.501642935377875,
+      "grad_norm": 1.0947457551956177,
+      "learning_rate": 0.0003386099657387468,
+      "loss": 0.6049380493164063,
+      "mean_token_accuracy": 0.818329308629036,
+      "num_tokens": 3641504.0,
+      "step": 1600
+    },
+    {
+      "entropy": 0.6244189664721489,
+      "epoch": 3.6111719605695507,
+      "grad_norm": 1.2528455257415771,
+      "learning_rate": 0.00033248757531473597,
+      "loss": 0.6117112731933594,
+      "mean_token_accuracy": 0.8165279686450958,
+      "num_tokens": 3752182.0,
+      "step": 1650
+    },
+    {
+      "entropy": 0.6266464674472809,
+      "epoch": 3.7207009857612268,
+      "grad_norm": 0.8912960290908813,
+      "learning_rate": 0.00032618083866292364,
+      "loss": 0.6233459854125977,
+      "mean_token_accuracy": 0.8161942487955094,
+      "num_tokens": 3866473.0,
+      "step": 1700
+    },
+    {
+      "entropy": 0.603082166314125,
+      "epoch": 3.8302300109529024,
+      "grad_norm": 0.9780993461608887,
+      "learning_rate": 0.00031969895337750556,
+      "loss": 0.5993933868408203,
+      "mean_token_accuracy": 0.8191226273775101,
+      "num_tokens": 3977393.0,
+      "step": 1750
+    },
+    {
+      "entropy": 0.6284996509552002,
+      "epoch": 3.9397590361445785,
+      "grad_norm": 1.3573766946792603,
+      "learning_rate": 0.00031305137248526394,
+      "loss": 0.6139850616455078,
+      "mean_token_accuracy": 0.8174162280559539,
+      "num_tokens": 4088936.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5731133787877976,
+      "eval_loss": 0.7714413404464722,
+      "eval_mean_token_accuracy": 0.7904208685670581,
+      "eval_num_tokens": 4153000.0,
+      "eval_runtime": 43.8857,
+      "eval_samples_per_second": 22.809,
+      "eval_steps_per_second": 2.871,
+      "step": 1828
+    },
+    {
+      "entropy": 0.56682777645612,
+      "epoch": 4.048192771084337,
+      "grad_norm": 0.9725717306137085,
+      "learning_rate": 0.0003062477906594989,
+      "loss": 0.5487574005126953,
+      "mean_token_accuracy": 0.8309493137128425,
+      "num_tokens": 4202329.0,
+      "step": 1850
+    },
+    {
+      "entropy": 0.522312518954277,
+      "epoch": 4.157721796276014,
+      "grad_norm": 1.0364148616790771,
+      "learning_rate": 0.00029929813008154865,
+      "loss": 0.5065541458129883,
+      "mean_token_accuracy": 0.8395609927177429,
+      "num_tokens": 4320321.0,
+      "step": 1900
+    },
+    {
+      "entropy": 0.5235522103309631,
+      "epoch": 4.267250821467689,
+      "grad_norm": 0.9119980931282043,
+      "learning_rate": 0.00029221252597051635,
+      "loss": 0.5116676712036132,
+      "mean_token_accuracy": 0.8383910119533539,
+      "num_tokens": 4435139.0,
+      "step": 1950
+    },
+    {
+      "entropy": 0.5248301169276237,
+      "epoch": 4.376779846659365,
+      "grad_norm": 1.2733819484710693,
+      "learning_rate": 0.00028500131180230935,
+      "loss": 0.5151065826416016,
+      "mean_token_accuracy": 0.8378184515237809,
+      "num_tokens": 4547272.0,
+      "step": 2000
+    },
+    {
+      "entropy": 0.531632961332798,
+      "epoch": 4.48630887185104,
+      "grad_norm": 1.008576512336731,
+      "learning_rate": 0.00027767500423954395,
+      "loss": 0.5181634902954102,
+      "mean_token_accuracy": 0.8388127386569977,
+      "num_tokens": 4659036.0,
+      "step": 2050
+    },
+    {
+      "entropy": 0.5213253930211067,
+      "epoch": 4.595837897042716,
+      "grad_norm": 0.9017001986503601,
+      "learning_rate": 0.0002702442877942959,
+      "loss": 0.5104235458374023,
+      "mean_token_accuracy": 0.8390659308433532,
+      "num_tokens": 4769957.0,
+      "step": 2100
+    },
+    {
+      "entropy": 0.5312498080730438,
+      "epoch": 4.705366922234392,
+      "grad_norm": 0.8270431756973267,
+      "learning_rate": 0.00026271999924606297,
+      "loss": 0.5256425476074219,
+      "mean_token_accuracy": 0.8363041788339615,
+      "num_tokens": 4882487.0,
+      "step": 2150
+    },
+    {
+      "entropy": 0.5297493544220925,
+      "epoch": 4.814895947426068,
+      "grad_norm": 0.6612098813056946,
+      "learning_rate": 0.00025511311183766457,
+      "loss": 0.5210869979858398,
+      "mean_token_accuracy": 0.837263793349266,
+      "num_tokens": 4993060.0,
+      "step": 2200
+    },
+    {
+      "entropy": 0.5250286677479744,
+      "epoch": 4.924424972617744,
+      "grad_norm": 0.8537449240684509,
+      "learning_rate": 0.00024743471927212595,
+      "loss": 0.5147652816772461,
+      "mean_token_accuracy": 0.8392492067813874,
+      "num_tokens": 5111182.0,
+      "step": 2250
+    },
+    {
+      "epoch": 5.0,
+      "eval_entropy": 0.5230902518544879,
+      "eval_loss": 0.7304916977882385,
+      "eval_mean_token_accuracy": 0.8025014537667471,
+      "eval_num_tokens": 5191250.0,
+      "eval_runtime": 43.9445,
+      "eval_samples_per_second": 22.779,
+      "eval_steps_per_second": 2.867,
+      "step": 2285
+    },
+    {
+      "entropy": 0.48566745266769873,
+      "epoch": 5.032858707557502,
+      "grad_norm": 0.7822101712226868,
+      "learning_rate": 0.000239696019533887,
+      "loss": 0.4735472869873047,
+      "mean_token_accuracy": 0.8504744623646592,
+      "num_tokens": 5225197.0,
+      "step": 2300
+    },
+    {
+      "entropy": 0.40831263691186903,
+      "epoch": 5.142387732749179,
+      "grad_norm": 0.8123568296432495,
+      "learning_rate": 0.00023190829855792867,
+      "loss": 0.3874293518066406,
+      "mean_token_accuracy": 0.8715608316659927,
+      "num_tokens": 5336728.0,
+      "step": 2350
+    },
+    {
+      "entropy": 0.41975140184164045,
+      "epoch": 5.2519167579408546,
+      "grad_norm": 0.7969071865081787,
+      "learning_rate": 0.00022408291377063483,
+      "loss": 0.40011211395263674,
+      "mean_token_accuracy": 0.8658375990390778,
+      "num_tokens": 5446986.0,
+      "step": 2400
+    },
+    {
+      "entropy": 0.41357740610837934,
+      "epoch": 5.36144578313253,
+      "grad_norm": 1.0098742246627808,
+      "learning_rate": 0.000216231277526393,
+      "loss": 0.39797264099121094,
+      "mean_token_accuracy": 0.8674204486608506,
+      "num_tokens": 5563479.0,
+      "step": 2450
+    },
+    {
+      "entropy": 0.4265018665790558,
+      "epoch": 5.470974808324206,
+      "grad_norm": 1.160515308380127,
+      "learning_rate": 0.00020836484046408868,
+      "loss": 0.40515335083007814,
+      "mean_token_accuracy": 0.8648038375377655,
+      "num_tokens": 5677338.0,
+      "step": 2500
+    },
+    {
+      "entropy": 0.4268675494194031,
+      "epoch": 5.580503833515881,
+      "grad_norm": 0.7267200946807861,
+      "learning_rate": 0.0002004950748077673,
+      "loss": 0.40761962890625,
+      "mean_token_accuracy": 0.864390144944191,
+      "num_tokens": 5787988.0,
+      "step": 2550
+    },
+    {
+      "entropy": 0.4280226981639862,
+      "epoch": 5.690032858707557,
+      "grad_norm": 0.8958321809768677,
+      "learning_rate": 0.00019263345763581628,
+      "loss": 0.40727943420410156,
+      "mean_token_accuracy": 0.8640619122982025,
+      "num_tokens": 5902559.0,
+      "step": 2600
+    },
+    {
+      "entropy": 0.431531577706337,
+      "epoch": 5.7995618838992335,
+      "grad_norm": 0.875791072845459,
+      "learning_rate": 0.00018479145414306803,
+      "loss": 0.40996681213378905,
+      "mean_token_accuracy": 0.8628572630882263,
+      "num_tokens": 6014039.0,
+      "step": 2650
+    },
+    {
+      "entropy": 0.41479706585407256,
+      "epoch": 5.909090909090909,
+      "grad_norm": 0.7954572439193726,
+      "learning_rate": 0.0001769805009202332,
+      "loss": 0.3963738250732422,
+      "mean_token_accuracy": 0.8696580404043197,
+      "num_tokens": 6134464.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.0,
+      "eval_entropy": 0.472232200796642,
+      "eval_loss": 0.747288703918457,
+      "eval_mean_token_accuracy": 0.802532631726492,
+      "eval_num_tokens": 6229500.0,
+      "eval_runtime": 43.8095,
+      "eval_samples_per_second": 22.849,
+      "eval_steps_per_second": 2.876,
+      "step": 2742
+    },
+    {
+      "entropy": 0.405825153143719,
+      "epoch": 6.017524644030668,
+      "grad_norm": 0.675167977809906,
+      "learning_rate": 0.00016921198927505012,
+      "loss": 0.38604618072509767,
+      "mean_token_accuracy": 0.8706694215235083,
+      "num_tokens": 6249905.0,
+      "step": 2750
+    },
+    {
+      "entropy": 0.28874062046408655,
+      "epoch": 6.127053669222344,
+      "grad_norm": 0.8683274984359741,
+      "learning_rate": 0.0001614972486194732,
+      "loss": 0.2581937789916992,
+      "mean_token_accuracy": 0.9093546283245086,
+      "num_tokens": 6367665.0,
+      "step": 2800
+    },
+    {
+      "entropy": 0.29525566384196283,
+      "epoch": 6.23658269441402,
+      "grad_norm": 0.8548017740249634,
+      "learning_rate": 0.0001538475299471291,
+      "loss": 0.26408302307128906,
+      "mean_token_accuracy": 0.9077397453784942,
+      "num_tokens": 6481769.0,
+      "step": 2850
+    },
+    {
+      "entropy": 0.30853822842240336,
+      "epoch": 6.3461117196056955,
+      "grad_norm": 0.9827961325645447,
+      "learning_rate": 0.00014627398942513635,
+      "loss": 0.2714972877502441,
+      "mean_token_accuracy": 0.9044403052330017,
+      "num_tokens": 6595712.0,
+      "step": 2900
+    },
+    {
+      "entropy": 0.30256781667470933,
+      "epoch": 6.455640744797371,
+      "grad_norm": 0.8727903962135315,
+      "learning_rate": 0.0001387876721242171,
+      "loss": 0.2691990852355957,
+      "mean_token_accuracy": 0.9049568754434586,
+      "num_tokens": 6708841.0,
+      "step": 2950
+    },
+    {
+      "entropy": 0.305324949324131,
+      "epoch": 6.565169769989047,
+      "grad_norm": 0.7359956502914429,
+      "learning_rate": 0.00013139949591082985,
+      "loss": 0.274547176361084,
+      "mean_token_accuracy": 0.9040277010202408,
+      "num_tokens": 6818025.0,
+      "step": 3000
+    },
+    {
+      "entropy": 0.3107963638007641,
+      "epoch": 6.674698795180722,
+      "grad_norm": 0.814873456954956,
+      "learning_rate": 0.0001241202355248137,
+      "loss": 0.27763776779174804,
+      "mean_token_accuracy": 0.9029175835847855,
+      "num_tokens": 6929746.0,
+      "step": 3050
+    },
+    {
+      "entropy": 0.30067121908068656,
+      "epoch": 6.784227820372399,
+      "grad_norm": 0.796090841293335,
+      "learning_rate": 0.00011696050686576498,
+      "loss": 0.26881502151489256,
+      "mean_token_accuracy": 0.9052424907684327,
+      "num_tokens": 7046109.0,
+      "step": 3100
+    },
+    {
+      "entropy": 0.30515122786164284,
+      "epoch": 6.8937568455640745,
+      "grad_norm": 0.8339642286300659,
+      "learning_rate": 0.00010993075151106208,
+      "loss": 0.273194522857666,
+      "mean_token_accuracy": 0.904237716794014,
+      "num_tokens": 7156155.0,
+      "step": 3150
+    },
+    {
+      "epoch": 7.0,
+      "eval_entropy": 0.37887449586202226,
+      "eval_loss": 0.8343665599822998,
+      "eval_mean_token_accuracy": 0.8046237868922097,
+      "eval_num_tokens": 7267750.0,
+      "eval_runtime": 43.2059,
+      "eval_samples_per_second": 23.168,
+      "eval_steps_per_second": 2.916,
+      "step": 3199
+    },
+    {
+      "entropy": 0.29947714342011345,
+      "epoch": 7.002190580503833,
+      "grad_norm": 0.5025448799133301,
+      "learning_rate": 0.0001030412214881193,
+      "loss": 0.2715941619873047,
+      "mean_token_accuracy": 0.9054891123916163,
+      "num_tokens": 7269424.0,
+      "step": 3200
+    },
+    {
+      "entropy": 0.19013839438557625,
+      "epoch": 7.11171960569551,
+      "grad_norm": 0.7953240275382996,
+      "learning_rate": 9.630196432307369e-05,
+      "loss": 0.15198980331420897,
+      "mean_token_accuracy": 0.9485088515281678,
+      "num_tokens": 7382588.0,
+      "step": 3250
+    },
+    {
+      "entropy": 0.18270175024867058,
+      "epoch": 7.221248630887185,
+      "grad_norm": 0.8207944631576538,
+      "learning_rate": 8.972280838771363e-05,
+      "loss": 0.1504285717010498,
+      "mean_token_accuracy": 0.9483412396907807,
+      "num_tokens": 7494048.0,
+      "step": 3300
+    },
+    {
+      "entropy": 0.1822182248532772,
+      "epoch": 7.330777656078861,
+      "grad_norm": 0.7845531105995178,
+      "learning_rate": 8.331334856601539e-05,
+      "loss": 0.1527143096923828,
+      "mean_token_accuracy": 0.9482884389162064,
+      "num_tokens": 7607343.0,
+      "step": 3350
+    },
+    {
+      "entropy": 0.18213839918375016,
+      "epoch": 7.4403066812705365,
+      "grad_norm": 0.9096264839172363,
+      "learning_rate": 7.708293226119355e-05,
+      "loss": 0.1504144287109375,
+      "mean_token_accuracy": 0.9482587432861328,
+      "num_tokens": 7724520.0,
+      "step": 3400
+    },
+    {
+      "entropy": 0.18474044188857078,
+      "epoch": 7.549835706462212,
+      "grad_norm": 0.7610820531845093,
+      "learning_rate": 7.104064576367165e-05,
+      "loss": 0.15584409713745118,
+      "mean_token_accuracy": 0.9465331947803497,
+      "num_tokens": 7836568.0,
+      "step": 3450
+    },
+    {
+      "entropy": 0.1816471180319786,
+      "epoch": 7.659364731653889,
+      "grad_norm": 0.8613850474357605,
+      "learning_rate": 6.519530099985234e-05,
+      "loss": 0.15603148460388183,
+      "mean_token_accuracy": 0.9455479162931443,
+      "num_tokens": 7947988.0,
+      "step": 3500
+    },
+    {
+      "entropy": 0.18248806044459343,
+      "epoch": 7.768893756845564,
+      "grad_norm": 0.644688069820404,
+      "learning_rate": 5.9555422681015177e-05,
+      "loss": 0.1541291046142578,
+      "mean_token_accuracy": 0.9463654214143753,
+      "num_tokens": 8066809.0,
+      "step": 3550
+    },
+    {
+      "entropy": 0.1824373207241297,
+      "epoch": 7.87842278203724,
+      "grad_norm": 0.7455086708068848,
+      "learning_rate": 5.4129235871080286e-05,
+      "loss": 0.15226722717285157,
+      "mean_token_accuracy": 0.9470921188592911,
+      "num_tokens": 8178429.0,
+      "step": 3600
+    },
+    {
+      "entropy": 0.17558989696204663,
+      "epoch": 7.9879518072289155,
+      "grad_norm": 0.6741284728050232,
+      "learning_rate": 4.8924653991372106e-05,
+      "loss": 0.14881887435913085,
+      "mean_token_accuracy": 0.9479702442884446,
+      "num_tokens": 8294498.0,
+      "step": 3650
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.31352401418345316,
+      "eval_loss": 0.9514623284339905,
+      "eval_mean_token_accuracy": 0.8020065157186418,
+      "eval_num_tokens": 8306000.0,
+      "eval_runtime": 43.9551,
+      "eval_samples_per_second": 22.773,
+      "eval_steps_per_second": 2.867,
+      "step": 3656
+    },
+    {
+      "entropy": 0.1254210752068144,
+      "epoch": 8.096385542168674,
+      "grad_norm": 0.4076255261898041,
+      "learning_rate": 4.394926727987422e-05,
+      "loss": 0.09436151504516602,
+      "mean_token_accuracy": 0.9698121258706758,
+      "num_tokens": 8408606.0,
+      "step": 3700
+    },
+    {
+      "entropy": 0.11122448101639748,
+      "epoch": 8.20591456736035,
+      "grad_norm": 0.42469334602355957,
+      "learning_rate": 3.921033172180787e-05,
+      "loss": 0.0847774600982666,
+      "mean_token_accuracy": 0.972111673951149,
+      "num_tokens": 8523042.0,
+      "step": 3750
+    },
+    {
+      "entropy": 0.1047579725086689,
+      "epoch": 8.315443592552027,
+      "grad_norm": 0.5445597767829895,
+      "learning_rate": 3.47147584676758e-05,
+      "loss": 0.08128767967224121,
+      "mean_token_accuracy": 0.9733622133731842,
+      "num_tokens": 8640478.0,
+      "step": 3800
+    },
+    {
+      "entropy": 0.11103030495345592,
+      "epoch": 8.424972617743702,
+      "grad_norm": 0.4622252881526947,
+      "learning_rate": 3.046910375420573e-05,
+      "loss": 0.08711739540100098,
+      "mean_token_accuracy": 0.9709586310386658,
+      "num_tokens": 8752683.0,
+      "step": 3850
+    },
+    {
+      "entropy": 0.10623676404356956,
+      "epoch": 8.534501642935378,
+      "grad_norm": 0.47963249683380127,
+      "learning_rate": 2.647955934289092e-05,
+      "loss": 0.08331462860107422,
+      "mean_token_accuracy": 0.9721733230352402,
+      "num_tokens": 8869555.0,
+      "step": 3900
+    },
+    {
+      "entropy": 0.1133701204881072,
+      "epoch": 8.644030668127053,
+      "grad_norm": 0.5320031642913818,
+      "learning_rate": 2.275194349007337e-05,
+      "loss": 0.08723929405212402,
+      "mean_token_accuracy": 0.9707023686170578,
+      "num_tokens": 8977869.0,
+      "step": 3950
+    },
+    {
+      "entropy": 0.11115138165652752,
+      "epoch": 8.75355969331873,
+      "grad_norm": 0.46053412556648254,
+      "learning_rate": 1.9291692461737163e-05,
+      "loss": 0.08680099487304688,
+      "mean_token_accuracy": 0.970695189833641,
+      "num_tokens": 9085599.0,
+      "step": 4000
+    },
+    {
+      "entropy": 0.10649645052850247,
+      "epoch": 8.863088718510404,
+      "grad_norm": 0.4637404978275299,
+      "learning_rate": 1.6103852605388375e-05,
+      "loss": 0.08539060592651367,
+      "mean_token_accuracy": 0.9716807705163956,
+      "num_tokens": 9196726.0,
+      "step": 4050
+    },
+    {
+      "entropy": 0.10305101312696933,
+      "epoch": 8.97261774370208,
+      "grad_norm": 0.4751088321208954,
+      "learning_rate": 1.319307299058198e-05,
+      "loss": 0.08058808326721191,
+      "mean_token_accuracy": 0.9734358698129654,
+      "num_tokens": 9314678.0,
+      "step": 4100
+    },
+    {
+      "epoch": 9.0,
+      "eval_entropy": 0.2646038674172901,
+      "eval_loss": 1.105322241783142,
+      "eval_mean_token_accuracy": 0.8011843391827175,
+      "eval_num_tokens": 9344250.0,
+      "eval_runtime": 44.0589,
+      "eval_samples_per_second": 22.72,
+      "eval_steps_per_second": 2.86,
+      "step": 4113
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 4570,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.0262519999913536e+17,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-457/README.md b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-457/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..7ac50e2706e5b4eae8f28e204601afb241c49e55
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-457/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-2B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-2B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-457/adapter_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-457/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..4d005ec8c649e4abb0922fac729d1444177e0dc1
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-457/adapter_config.json
@@ -0,0 +1,48 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.0029143607590526367,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "v_proj",
+    "o_proj",
+    "k_proj",
+    "up_proj",
+    "q_proj",
+    "down_proj",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-457/chat_template.jinja b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-457/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..0ef09f214eaa6d9bca297988afc1454b5827b2c7
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-457/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is true %}
+        {{- '<think>\n' }}
+    {%- else %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-457/tokenizer_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-457/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-457/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-457/trainer_state.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-457/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..8741244a7cfa4833e0b5faa36d54480dd2bafdda
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-457/trainer_state.json
@@ -0,0 +1,135 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.0,
+  "eval_steps": 500,
+  "global_step": 457,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 1.664013209939003,
+      "epoch": 0.10952902519167579,
+      "grad_norm": 1.825055480003357,
+      "learning_rate": 4.4192733942388133e-05,
+      "loss": 1.5422621154785157,
+      "mean_token_accuracy": 0.6574381196498871,
+      "num_tokens": 113087.0,
+      "step": 50
+    },
+    {
+      "entropy": 0.9298514115810395,
+      "epoch": 0.21905805038335158,
+      "grad_norm": 1.4702502489089966,
+      "learning_rate": 8.928736041421276e-05,
+      "loss": 0.8554936218261718,
+      "mean_token_accuracy": 0.7638577824831009,
+      "num_tokens": 230392.0,
+      "step": 100
+    },
+    {
+      "entropy": 0.8588321554660797,
+      "epoch": 0.32858707557502737,
+      "grad_norm": 1.1502039432525635,
+      "learning_rate": 0.00013438198688603737,
+      "loss": 0.7965679168701172,
+      "mean_token_accuracy": 0.7794932717084885,
+      "num_tokens": 341746.0,
+      "step": 150
+    },
+    {
+      "entropy": 0.8281867235898972,
+      "epoch": 0.43811610076670315,
+      "grad_norm": 1.3328826427459717,
+      "learning_rate": 0.00017947661335786198,
+      "loss": 0.7692495727539063,
+      "mean_token_accuracy": 0.7842007935047149,
+      "num_tokens": 460483.0,
+      "step": 200
+    },
+    {
+      "entropy": 0.8131981420516968,
+      "epoch": 0.547645125958379,
+      "grad_norm": 1.2384045124053955,
+      "learning_rate": 0.00022457123982968662,
+      "loss": 0.7567300415039062,
+      "mean_token_accuracy": 0.7894466006755829,
+      "num_tokens": 570627.0,
+      "step": 250
+    },
+    {
+      "entropy": 0.8158841520547867,
+      "epoch": 0.6571741511500547,
+      "grad_norm": 1.29042387008667,
+      "learning_rate": 0.00026966586630151125,
+      "loss": 0.762672119140625,
+      "mean_token_accuracy": 0.7883876228332519,
+      "num_tokens": 677217.0,
+      "step": 300
+    },
+    {
+      "entropy": 0.8154043215513229,
+      "epoch": 0.7667031763417306,
+      "grad_norm": 1.4341716766357422,
+      "learning_rate": 0.00031476049277333584,
+      "loss": 0.7649243927001953,
+      "mean_token_accuracy": 0.7877721953392028,
+      "num_tokens": 790185.0,
+      "step": 350
+    },
+    {
+      "entropy": 0.8017062950134277,
+      "epoch": 0.8762322015334063,
+      "grad_norm": 1.2653132677078247,
+      "learning_rate": 0.0003598551192451605,
+      "loss": 0.7603777313232422,
+      "mean_token_accuracy": 0.7864453828334809,
+      "num_tokens": 910080.0,
+      "step": 400
+    },
+    {
+      "entropy": 0.8266482776403428,
+      "epoch": 0.9857612267250822,
+      "grad_norm": 1.1987895965576172,
+      "learning_rate": 0.0004049497457169851,
+      "loss": 0.783993911743164,
+      "mean_token_accuracy": 0.7832800775766373,
+      "num_tokens": 1023473.0,
+      "step": 450
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.8228033265424153,
+      "eval_loss": 0.8296912908554077,
+      "eval_mean_token_accuracy": 0.769342744634265,
+      "eval_num_tokens": 1038250.0,
+      "eval_runtime": 44.5194,
+      "eval_samples_per_second": 22.485,
+      "eval_steps_per_second": 2.83,
+      "step": 457
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 4570,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.245682367536256e+16,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4570/README.md b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4570/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..7ac50e2706e5b4eae8f28e204601afb241c49e55
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4570/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-2B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-2B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4570/adapter_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4570/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..4d005ec8c649e4abb0922fac729d1444177e0dc1
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4570/adapter_config.json
@@ -0,0 +1,48 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.0029143607590526367,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "v_proj",
+    "o_proj",
+    "k_proj",
+    "up_proj",
+    "q_proj",
+    "down_proj",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4570/chat_template.jinja b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4570/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..0ef09f214eaa6d9bca297988afc1454b5827b2c7
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4570/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is true %}
+        {{- '<think>\n' }}
+    {%- else %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4570/tokenizer_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4570/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4570/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4570/trainer_state.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4570/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..b9d8add6d9c7d9357082fa38b3831da56ec8eb6a
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-4570/trainer_state.json
@@ -0,0 +1,1054 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 10.0,
+  "eval_steps": 500,
+  "global_step": 4570,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 1.664013209939003,
+      "epoch": 0.10952902519167579,
+      "grad_norm": 1.825055480003357,
+      "learning_rate": 4.4192733942388133e-05,
+      "loss": 1.5422621154785157,
+      "mean_token_accuracy": 0.6574381196498871,
+      "num_tokens": 113087.0,
+      "step": 50
+    },
+    {
+      "entropy": 0.9298514115810395,
+      "epoch": 0.21905805038335158,
+      "grad_norm": 1.4702502489089966,
+      "learning_rate": 8.928736041421276e-05,
+      "loss": 0.8554936218261718,
+      "mean_token_accuracy": 0.7638577824831009,
+      "num_tokens": 230392.0,
+      "step": 100
+    },
+    {
+      "entropy": 0.8588321554660797,
+      "epoch": 0.32858707557502737,
+      "grad_norm": 1.1502039432525635,
+      "learning_rate": 0.00013438198688603737,
+      "loss": 0.7965679168701172,
+      "mean_token_accuracy": 0.7794932717084885,
+      "num_tokens": 341746.0,
+      "step": 150
+    },
+    {
+      "entropy": 0.8281867235898972,
+      "epoch": 0.43811610076670315,
+      "grad_norm": 1.3328826427459717,
+      "learning_rate": 0.00017947661335786198,
+      "loss": 0.7692495727539063,
+      "mean_token_accuracy": 0.7842007935047149,
+      "num_tokens": 460483.0,
+      "step": 200
+    },
+    {
+      "entropy": 0.8131981420516968,
+      "epoch": 0.547645125958379,
+      "grad_norm": 1.2384045124053955,
+      "learning_rate": 0.00022457123982968662,
+      "loss": 0.7567300415039062,
+      "mean_token_accuracy": 0.7894466006755829,
+      "num_tokens": 570627.0,
+      "step": 250
+    },
+    {
+      "entropy": 0.8158841520547867,
+      "epoch": 0.6571741511500547,
+      "grad_norm": 1.29042387008667,
+      "learning_rate": 0.00026966586630151125,
+      "loss": 0.762672119140625,
+      "mean_token_accuracy": 0.7883876228332519,
+      "num_tokens": 677217.0,
+      "step": 300
+    },
+    {
+      "entropy": 0.8154043215513229,
+      "epoch": 0.7667031763417306,
+      "grad_norm": 1.4341716766357422,
+      "learning_rate": 0.00031476049277333584,
+      "loss": 0.7649243927001953,
+      "mean_token_accuracy": 0.7877721953392028,
+      "num_tokens": 790185.0,
+      "step": 350
+    },
+    {
+      "entropy": 0.8017062950134277,
+      "epoch": 0.8762322015334063,
+      "grad_norm": 1.2653132677078247,
+      "learning_rate": 0.0003598551192451605,
+      "loss": 0.7603777313232422,
+      "mean_token_accuracy": 0.7864453828334809,
+      "num_tokens": 910080.0,
+      "step": 400
+    },
+    {
+      "entropy": 0.8266482776403428,
+      "epoch": 0.9857612267250822,
+      "grad_norm": 1.1987895965576172,
+      "learning_rate": 0.0004049497457169851,
+      "loss": 0.783993911743164,
+      "mean_token_accuracy": 0.7832800775766373,
+      "num_tokens": 1023473.0,
+      "step": 450
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.8228033265424153,
+      "eval_loss": 0.8296912908554077,
+      "eval_mean_token_accuracy": 0.769342744634265,
+      "eval_num_tokens": 1038250.0,
+      "eval_runtime": 44.5194,
+      "eval_samples_per_second": 22.485,
+      "eval_steps_per_second": 2.83,
+      "step": 457
+    },
+    {
+      "entropy": 0.8000734815693865,
+      "epoch": 1.0941949616648412,
+      "grad_norm": 1.624888300895691,
+      "learning_rate": 0.00041205884962961745,
+      "loss": 0.7572084045410157,
+      "mean_token_accuracy": 0.7857678087070735,
+      "num_tokens": 1128922.0,
+      "step": 500
+    },
+    {
+      "entropy": 0.7889713329076767,
+      "epoch": 1.203723986856517,
+      "grad_norm": 1.5562400817871094,
+      "learning_rate": 0.0004116562696726163,
+      "loss": 0.757301025390625,
+      "mean_token_accuracy": 0.7859748858213425,
+      "num_tokens": 1245635.0,
+      "step": 550
+    },
+    {
+      "entropy": 0.7770237320661545,
+      "epoch": 1.3132530120481927,
+      "grad_norm": 1.7213724851608276,
+      "learning_rate": 0.00041095388575366294,
+      "loss": 0.7583953857421875,
+      "mean_token_accuracy": 0.7878059667348861,
+      "num_tokens": 1354885.0,
+      "step": 600
+    },
+    {
+      "entropy": 0.7724025696516037,
+      "epoch": 1.4227820372398685,
+      "grad_norm": 1.2281577587127686,
+      "learning_rate": 0.00040995272221269925,
+      "loss": 0.7478806304931641,
+      "mean_token_accuracy": 0.789793593287468,
+      "num_tokens": 1472232.0,
+      "step": 650
+    },
+    {
+      "entropy": 0.7704478764533996,
+      "epoch": 1.5323110624315444,
+      "grad_norm": 1.5951343774795532,
+      "learning_rate": 0.00040865423912273625,
+      "loss": 0.7493467712402344,
+      "mean_token_accuracy": 0.7901222252845764,
+      "num_tokens": 1587494.0,
+      "step": 700
+    },
+    {
+      "entropy": 0.7755234348773956,
+      "epoch": 1.6418400876232202,
+      "grad_norm": 1.2122973203659058,
+      "learning_rate": 0.0004070603301605186,
+      "loss": 0.7549277496337891,
+      "mean_token_accuracy": 0.7899289625883102,
+      "num_tokens": 1696927.0,
+      "step": 750
+    },
+    {
+      "entropy": 0.7561545193195343,
+      "epoch": 1.751369112814896,
+      "grad_norm": 1.262399673461914,
+      "learning_rate": 0.00040517331984483133,
+      "loss": 0.7413782501220703,
+      "mean_token_accuracy": 0.7917980045080185,
+      "num_tokens": 1813415.0,
+      "step": 800
+    },
+    {
+      "entropy": 0.7466552203893662,
+      "epoch": 1.8608981380065717,
+      "grad_norm": 1.2854546308517456,
+      "learning_rate": 0.00040299596014647715,
+      "loss": 0.7443357849121094,
+      "mean_token_accuracy": 0.79294473528862,
+      "num_tokens": 1924911.0,
+      "step": 850
+    },
+    {
+      "entropy": 0.7460890018939972,
+      "epoch": 1.9704271631982475,
+      "grad_norm": 1.0535873174667358,
+      "learning_rate": 0.00040053142647486815,
+      "loss": 0.742437973022461,
+      "mean_token_accuracy": 0.7911104756593704,
+      "num_tokens": 2044650.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.0,
+      "eval_entropy": 0.7546336958332668,
+      "eval_loss": 0.7908218502998352,
+      "eval_mean_token_accuracy": 0.7816079741432553,
+      "eval_num_tokens": 2076500.0,
+      "eval_runtime": 43.6577,
+      "eval_samples_per_second": 22.928,
+      "eval_steps_per_second": 2.886,
+      "step": 914
+    },
+    {
+      "entropy": 0.6920583218035071,
+      "epoch": 2.078860898138007,
+      "grad_norm": 0.8588879704475403,
+      "learning_rate": 0.0003977833130470843,
+      "loss": 0.6903114318847656,
+      "mean_token_accuracy": 0.7997718549738027,
+      "num_tokens": 2160498.0,
+      "step": 950
+    },
+    {
+      "entropy": 0.6784634017944335,
+      "epoch": 2.1883899233296824,
+      "grad_norm": 1.1033616065979004,
+      "learning_rate": 0.0003947556276461532,
+      "loss": 0.6780696868896484,
+      "mean_token_accuracy": 0.8036011070013046,
+      "num_tokens": 2269979.0,
+      "step": 1000
+    },
+    {
+      "entropy": 0.6898973470926285,
+      "epoch": 2.297918948521358,
+      "grad_norm": 1.1214436292648315,
+      "learning_rate": 0.0003914527857761949,
+      "loss": 0.6930846405029297,
+      "mean_token_accuracy": 0.8000095456838607,
+      "num_tokens": 2378362.0,
+      "step": 1050
+    },
+    {
+      "entropy": 0.6754742419719696,
+      "epoch": 2.407447973713034,
+      "grad_norm": 1.0675715208053589,
+      "learning_rate": 0.0003878796042229559,
+      "loss": 0.6745318603515625,
+      "mean_token_accuracy": 0.8050142484903335,
+      "num_tokens": 2498494.0,
+      "step": 1100
+    },
+    {
+      "entropy": 0.6677694636583328,
+      "epoch": 2.5169769989047097,
+      "grad_norm": 1.7671644687652588,
+      "learning_rate": 0.00038404129402912424,
+      "loss": 0.6703590393066406,
+      "mean_token_accuracy": 0.8045552480220795,
+      "num_tokens": 2611646.0,
+      "step": 1150
+    },
+    {
+      "entropy": 0.6750778061151504,
+      "epoch": 2.6265060240963853,
+      "grad_norm": 1.428247332572937,
+      "learning_rate": 0.00037994345289466886,
+      "loss": 0.6790786743164062,
+      "mean_token_accuracy": 0.8028086322546005,
+      "num_tokens": 2717744.0,
+      "step": 1200
+    },
+    {
+      "entropy": 0.6834373778104782,
+      "epoch": 2.7360350492880614,
+      "grad_norm": 0.9393771886825562,
+      "learning_rate": 0.0003755920570132876,
+      "loss": 0.6844132232666016,
+      "mean_token_accuracy": 0.8027357232570648,
+      "num_tokens": 2827750.0,
+      "step": 1250
+    },
+    {
+      "entropy": 0.671256046295166,
+      "epoch": 2.845564074479737,
+      "grad_norm": 1.2709442377090454,
+      "learning_rate": 0.0003709934523568692,
+      "loss": 0.6658607482910156,
+      "mean_token_accuracy": 0.8065212601423264,
+      "num_tokens": 2948958.0,
+      "step": 1300
+    },
+    {
+      "entropy": 0.6620459565520287,
+      "epoch": 2.955093099671413,
+      "grad_norm": 1.0260528326034546,
+      "learning_rate": 0.00036615434542067925,
+      "loss": 0.6596417236328125,
+      "mean_token_accuracy": 0.8089053362607956,
+      "num_tokens": 3066917.0,
+      "step": 1350
+    },
+    {
+      "epoch": 3.0,
+      "eval_entropy": 0.6772078592625875,
+      "eval_loss": 0.7904035449028015,
+      "eval_mean_token_accuracy": 0.7809106636615026,
+      "eval_num_tokens": 3114750.0,
+      "eval_runtime": 44.0158,
+      "eval_samples_per_second": 22.742,
+      "eval_steps_per_second": 2.863,
+      "step": 1371
+    },
+    {
+      "entropy": 0.6129003967901673,
+      "epoch": 3.063526834611172,
+      "grad_norm": 1.0416573286056519,
+      "learning_rate": 0.00036108179344276724,
+      "loss": 0.6105740356445313,
+      "mean_token_accuracy": 0.8168958028157552,
+      "num_tokens": 3180366.0,
+      "step": 1400
+    },
+    {
+      "entropy": 0.6044845461845398,
+      "epoch": 3.1730558598028478,
+      "grad_norm": 0.9305349588394165,
+      "learning_rate": 0.00035578319411185954,
+      "loss": 0.59905029296875,
+      "mean_token_accuracy": 0.8190373265743256,
+      "num_tokens": 3293918.0,
+      "step": 1450
+    },
+    {
+      "entropy": 0.6040669977664948,
+      "epoch": 3.2825848849945234,
+      "grad_norm": 1.039084553718567,
+      "learning_rate": 0.00035026627477874664,
+      "loss": 0.5956837463378907,
+      "mean_token_accuracy": 0.8208023190498352,
+      "num_tokens": 3409058.0,
+      "step": 1500
+    },
+    {
+      "entropy": 0.5916981258988381,
+      "epoch": 3.3921139101861995,
+      "grad_norm": 1.1696960926055908,
+      "learning_rate": 0.00034453908118689904,
+      "loss": 0.5919752502441407,
+      "mean_token_accuracy": 0.8211179333925247,
+      "num_tokens": 3525131.0,
+      "step": 1550
+    },
+    {
+      "entropy": 0.6082543370127678,
+      "epoch": 3.501642935377875,
+      "grad_norm": 1.0947457551956177,
+      "learning_rate": 0.0003386099657387468,
+      "loss": 0.6049380493164063,
+      "mean_token_accuracy": 0.818329308629036,
+      "num_tokens": 3641504.0,
+      "step": 1600
+    },
+    {
+      "entropy": 0.6244189664721489,
+      "epoch": 3.6111719605695507,
+      "grad_norm": 1.2528455257415771,
+      "learning_rate": 0.00033248757531473597,
+      "loss": 0.6117112731933594,
+      "mean_token_accuracy": 0.8165279686450958,
+      "num_tokens": 3752182.0,
+      "step": 1650
+    },
+    {
+      "entropy": 0.6266464674472809,
+      "epoch": 3.7207009857612268,
+      "grad_norm": 0.8912960290908813,
+      "learning_rate": 0.00032618083866292364,
+      "loss": 0.6233459854125977,
+      "mean_token_accuracy": 0.8161942487955094,
+      "num_tokens": 3866473.0,
+      "step": 1700
+    },
+    {
+      "entropy": 0.603082166314125,
+      "epoch": 3.8302300109529024,
+      "grad_norm": 0.9780993461608887,
+      "learning_rate": 0.00031969895337750556,
+      "loss": 0.5993933868408203,
+      "mean_token_accuracy": 0.8191226273775101,
+      "num_tokens": 3977393.0,
+      "step": 1750
+    },
+    {
+      "entropy": 0.6284996509552002,
+      "epoch": 3.9397590361445785,
+      "grad_norm": 1.3573766946792603,
+      "learning_rate": 0.00031305137248526394,
+      "loss": 0.6139850616455078,
+      "mean_token_accuracy": 0.8174162280559539,
+      "num_tokens": 4088936.0,
+      "step": 1800
+    },
+    {
+      "epoch": 4.0,
+      "eval_entropy": 0.5731133787877976,
+      "eval_loss": 0.7714413404464722,
+      "eval_mean_token_accuracy": 0.7904208685670581,
+      "eval_num_tokens": 4153000.0,
+      "eval_runtime": 43.8857,
+      "eval_samples_per_second": 22.809,
+      "eval_steps_per_second": 2.871,
+      "step": 1828
+    },
+    {
+      "entropy": 0.56682777645612,
+      "epoch": 4.048192771084337,
+      "grad_norm": 0.9725717306137085,
+      "learning_rate": 0.0003062477906594989,
+      "loss": 0.5487574005126953,
+      "mean_token_accuracy": 0.8309493137128425,
+      "num_tokens": 4202329.0,
+      "step": 1850
+    },
+    {
+      "entropy": 0.522312518954277,
+      "epoch": 4.157721796276014,
+      "grad_norm": 1.0364148616790771,
+      "learning_rate": 0.00029929813008154865,
+      "loss": 0.5065541458129883,
+      "mean_token_accuracy": 0.8395609927177429,
+      "num_tokens": 4320321.0,
+      "step": 1900
+    },
+    {
+      "entropy": 0.5235522103309631,
+      "epoch": 4.267250821467689,
+      "grad_norm": 0.9119980931282043,
+      "learning_rate": 0.00029221252597051635,
+      "loss": 0.5116676712036132,
+      "mean_token_accuracy": 0.8383910119533539,
+      "num_tokens": 4435139.0,
+      "step": 1950
+    },
+    {
+      "entropy": 0.5248301169276237,
+      "epoch": 4.376779846659365,
+      "grad_norm": 1.2733819484710693,
+      "learning_rate": 0.00028500131180230935,
+      "loss": 0.5151065826416016,
+      "mean_token_accuracy": 0.8378184515237809,
+      "num_tokens": 4547272.0,
+      "step": 2000
+    },
+    {
+      "entropy": 0.531632961332798,
+      "epoch": 4.48630887185104,
+      "grad_norm": 1.008576512336731,
+      "learning_rate": 0.00027767500423954395,
+      "loss": 0.5181634902954102,
+      "mean_token_accuracy": 0.8388127386569977,
+      "num_tokens": 4659036.0,
+      "step": 2050
+    },
+    {
+      "entropy": 0.5213253930211067,
+      "epoch": 4.595837897042716,
+      "grad_norm": 0.9017001986503601,
+      "learning_rate": 0.0002702442877942959,
+      "loss": 0.5104235458374023,
+      "mean_token_accuracy": 0.8390659308433532,
+      "num_tokens": 4769957.0,
+      "step": 2100
+    },
+    {
+      "entropy": 0.5312498080730438,
+      "epoch": 4.705366922234392,
+      "grad_norm": 0.8270431756973267,
+      "learning_rate": 0.00026271999924606297,
+      "loss": 0.5256425476074219,
+      "mean_token_accuracy": 0.8363041788339615,
+      "num_tokens": 4882487.0,
+      "step": 2150
+    },
+    {
+      "entropy": 0.5297493544220925,
+      "epoch": 4.814895947426068,
+      "grad_norm": 0.6612098813056946,
+      "learning_rate": 0.00025511311183766457,
+      "loss": 0.5210869979858398,
+      "mean_token_accuracy": 0.837263793349266,
+      "num_tokens": 4993060.0,
+      "step": 2200
+    },
+    {
+      "entropy": 0.5250286677479744,
+      "epoch": 4.924424972617744,
+      "grad_norm": 0.8537449240684509,
+      "learning_rate": 0.00024743471927212595,
+      "loss": 0.5147652816772461,
+      "mean_token_accuracy": 0.8392492067813874,
+      "num_tokens": 5111182.0,
+      "step": 2250
+    },
+    {
+      "epoch": 5.0,
+      "eval_entropy": 0.5230902518544879,
+      "eval_loss": 0.7304916977882385,
+      "eval_mean_token_accuracy": 0.8025014537667471,
+      "eval_num_tokens": 5191250.0,
+      "eval_runtime": 43.9445,
+      "eval_samples_per_second": 22.779,
+      "eval_steps_per_second": 2.867,
+      "step": 2285
+    },
+    {
+      "entropy": 0.48566745266769873,
+      "epoch": 5.032858707557502,
+      "grad_norm": 0.7822101712226868,
+      "learning_rate": 0.000239696019533887,
+      "loss": 0.4735472869873047,
+      "mean_token_accuracy": 0.8504744623646592,
+      "num_tokens": 5225197.0,
+      "step": 2300
+    },
+    {
+      "entropy": 0.40831263691186903,
+      "epoch": 5.142387732749179,
+      "grad_norm": 0.8123568296432495,
+      "learning_rate": 0.00023190829855792867,
+      "loss": 0.3874293518066406,
+      "mean_token_accuracy": 0.8715608316659927,
+      "num_tokens": 5336728.0,
+      "step": 2350
+    },
+    {
+      "entropy": 0.41975140184164045,
+      "epoch": 5.2519167579408546,
+      "grad_norm": 0.7969071865081787,
+      "learning_rate": 0.00022408291377063483,
+      "loss": 0.40011211395263674,
+      "mean_token_accuracy": 0.8658375990390778,
+      "num_tokens": 5446986.0,
+      "step": 2400
+    },
+    {
+      "entropy": 0.41357740610837934,
+      "epoch": 5.36144578313253,
+      "grad_norm": 1.0098742246627808,
+      "learning_rate": 0.000216231277526393,
+      "loss": 0.39797264099121094,
+      "mean_token_accuracy": 0.8674204486608506,
+      "num_tokens": 5563479.0,
+      "step": 2450
+    },
+    {
+      "entropy": 0.4265018665790558,
+      "epoch": 5.470974808324206,
+      "grad_norm": 1.160515308380127,
+      "learning_rate": 0.00020836484046408868,
+      "loss": 0.40515335083007814,
+      "mean_token_accuracy": 0.8648038375377655,
+      "num_tokens": 5677338.0,
+      "step": 2500
+    },
+    {
+      "entropy": 0.4268675494194031,
+      "epoch": 5.580503833515881,
+      "grad_norm": 0.7267200946807861,
+      "learning_rate": 0.0002004950748077673,
+      "loss": 0.40761962890625,
+      "mean_token_accuracy": 0.864390144944191,
+      "num_tokens": 5787988.0,
+      "step": 2550
+    },
+    {
+      "entropy": 0.4280226981639862,
+      "epoch": 5.690032858707557,
+      "grad_norm": 0.8958321809768677,
+      "learning_rate": 0.00019263345763581628,
+      "loss": 0.40727943420410156,
+      "mean_token_accuracy": 0.8640619122982025,
+      "num_tokens": 5902559.0,
+      "step": 2600
+    },
+    {
+      "entropy": 0.431531577706337,
+      "epoch": 5.7995618838992335,
+      "grad_norm": 0.875791072845459,
+      "learning_rate": 0.00018479145414306803,
+      "loss": 0.40996681213378905,
+      "mean_token_accuracy": 0.8628572630882263,
+      "num_tokens": 6014039.0,
+      "step": 2650
+    },
+    {
+      "entropy": 0.41479706585407256,
+      "epoch": 5.909090909090909,
+      "grad_norm": 0.7954572439193726,
+      "learning_rate": 0.0001769805009202332,
+      "loss": 0.3963738250732422,
+      "mean_token_accuracy": 0.8696580404043197,
+      "num_tokens": 6134464.0,
+      "step": 2700
+    },
+    {
+      "epoch": 6.0,
+      "eval_entropy": 0.472232200796642,
+      "eval_loss": 0.747288703918457,
+      "eval_mean_token_accuracy": 0.802532631726492,
+      "eval_num_tokens": 6229500.0,
+      "eval_runtime": 43.8095,
+      "eval_samples_per_second": 22.849,
+      "eval_steps_per_second": 2.876,
+      "step": 2742
+    },
+    {
+      "entropy": 0.405825153143719,
+      "epoch": 6.017524644030668,
+      "grad_norm": 0.675167977809906,
+      "learning_rate": 0.00016921198927505012,
+      "loss": 0.38604618072509767,
+      "mean_token_accuracy": 0.8706694215235083,
+      "num_tokens": 6249905.0,
+      "step": 2750
+    },
+    {
+      "entropy": 0.28874062046408655,
+      "epoch": 6.127053669222344,
+      "grad_norm": 0.8683274984359741,
+      "learning_rate": 0.0001614972486194732,
+      "loss": 0.2581937789916992,
+      "mean_token_accuracy": 0.9093546283245086,
+      "num_tokens": 6367665.0,
+      "step": 2800
+    },
+    {
+      "entropy": 0.29525566384196283,
+      "epoch": 6.23658269441402,
+      "grad_norm": 0.8548017740249634,
+      "learning_rate": 0.0001538475299471291,
+      "loss": 0.26408302307128906,
+      "mean_token_accuracy": 0.9077397453784942,
+      "num_tokens": 6481769.0,
+      "step": 2850
+    },
+    {
+      "entropy": 0.30853822842240336,
+      "epoch": 6.3461117196056955,
+      "grad_norm": 0.9827961325645447,
+      "learning_rate": 0.00014627398942513635,
+      "loss": 0.2714972877502441,
+      "mean_token_accuracy": 0.9044403052330017,
+      "num_tokens": 6595712.0,
+      "step": 2900
+    },
+    {
+      "entropy": 0.30256781667470933,
+      "epoch": 6.455640744797371,
+      "grad_norm": 0.8727903962135315,
+      "learning_rate": 0.0001387876721242171,
+      "loss": 0.2691990852355957,
+      "mean_token_accuracy": 0.9049568754434586,
+      "num_tokens": 6708841.0,
+      "step": 2950
+    },
+    {
+      "entropy": 0.305324949324131,
+      "epoch": 6.565169769989047,
+      "grad_norm": 0.7359956502914429,
+      "learning_rate": 0.00013139949591082985,
+      "loss": 0.274547176361084,
+      "mean_token_accuracy": 0.9040277010202408,
+      "num_tokens": 6818025.0,
+      "step": 3000
+    },
+    {
+      "entropy": 0.3107963638007641,
+      "epoch": 6.674698795180722,
+      "grad_norm": 0.814873456954956,
+      "learning_rate": 0.0001241202355248137,
+      "loss": 0.27763776779174804,
+      "mean_token_accuracy": 0.9029175835847855,
+      "num_tokens": 6929746.0,
+      "step": 3050
+    },
+    {
+      "entropy": 0.30067121908068656,
+      "epoch": 6.784227820372399,
+      "grad_norm": 0.796090841293335,
+      "learning_rate": 0.00011696050686576498,
+      "loss": 0.26881502151489256,
+      "mean_token_accuracy": 0.9052424907684327,
+      "num_tokens": 7046109.0,
+      "step": 3100
+    },
+    {
+      "entropy": 0.30515122786164284,
+      "epoch": 6.8937568455640745,
+      "grad_norm": 0.8339642286300659,
+      "learning_rate": 0.00010993075151106208,
+      "loss": 0.273194522857666,
+      "mean_token_accuracy": 0.904237716794014,
+      "num_tokens": 7156155.0,
+      "step": 3150
+    },
+    {
+      "epoch": 7.0,
+      "eval_entropy": 0.37887449586202226,
+      "eval_loss": 0.8343665599822998,
+      "eval_mean_token_accuracy": 0.8046237868922097,
+      "eval_num_tokens": 7267750.0,
+      "eval_runtime": 43.2059,
+      "eval_samples_per_second": 23.168,
+      "eval_steps_per_second": 2.916,
+      "step": 3199
+    },
+    {
+      "entropy": 0.29947714342011345,
+      "epoch": 7.002190580503833,
+      "grad_norm": 0.5025448799133301,
+      "learning_rate": 0.0001030412214881193,
+      "loss": 0.2715941619873047,
+      "mean_token_accuracy": 0.9054891123916163,
+      "num_tokens": 7269424.0,
+      "step": 3200
+    },
+    {
+      "entropy": 0.19013839438557625,
+      "epoch": 7.11171960569551,
+      "grad_norm": 0.7953240275382996,
+      "learning_rate": 9.630196432307369e-05,
+      "loss": 0.15198980331420897,
+      "mean_token_accuracy": 0.9485088515281678,
+      "num_tokens": 7382588.0,
+      "step": 3250
+    },
+    {
+      "entropy": 0.18270175024867058,
+      "epoch": 7.221248630887185,
+      "grad_norm": 0.8207944631576538,
+      "learning_rate": 8.972280838771363e-05,
+      "loss": 0.1504285717010498,
+      "mean_token_accuracy": 0.9483412396907807,
+      "num_tokens": 7494048.0,
+      "step": 3300
+    },
+    {
+      "entropy": 0.1822182248532772,
+      "epoch": 7.330777656078861,
+      "grad_norm": 0.7845531105995178,
+      "learning_rate": 8.331334856601539e-05,
+      "loss": 0.1527143096923828,
+      "mean_token_accuracy": 0.9482884389162064,
+      "num_tokens": 7607343.0,
+      "step": 3350
+    },
+    {
+      "entropy": 0.18213839918375016,
+      "epoch": 7.4403066812705365,
+      "grad_norm": 0.9096264839172363,
+      "learning_rate": 7.708293226119355e-05,
+      "loss": 0.1504144287109375,
+      "mean_token_accuracy": 0.9482587432861328,
+      "num_tokens": 7724520.0,
+      "step": 3400
+    },
+    {
+      "entropy": 0.18474044188857078,
+      "epoch": 7.549835706462212,
+      "grad_norm": 0.7610820531845093,
+      "learning_rate": 7.104064576367165e-05,
+      "loss": 0.15584409713745118,
+      "mean_token_accuracy": 0.9465331947803497,
+      "num_tokens": 7836568.0,
+      "step": 3450
+    },
+    {
+      "entropy": 0.1816471180319786,
+      "epoch": 7.659364731653889,
+      "grad_norm": 0.8613850474357605,
+      "learning_rate": 6.519530099985234e-05,
+      "loss": 0.15603148460388183,
+      "mean_token_accuracy": 0.9455479162931443,
+      "num_tokens": 7947988.0,
+      "step": 3500
+    },
+    {
+      "entropy": 0.18248806044459343,
+      "epoch": 7.768893756845564,
+      "grad_norm": 0.644688069820404,
+      "learning_rate": 5.9555422681015177e-05,
+      "loss": 0.1541291046142578,
+      "mean_token_accuracy": 0.9463654214143753,
+      "num_tokens": 8066809.0,
+      "step": 3550
+    },
+    {
+      "entropy": 0.1824373207241297,
+      "epoch": 7.87842278203724,
+      "grad_norm": 0.7455086708068848,
+      "learning_rate": 5.4129235871080286e-05,
+      "loss": 0.15226722717285157,
+      "mean_token_accuracy": 0.9470921188592911,
+      "num_tokens": 8178429.0,
+      "step": 3600
+    },
+    {
+      "entropy": 0.17558989696204663,
+      "epoch": 7.9879518072289155,
+      "grad_norm": 0.6741284728050232,
+      "learning_rate": 4.8924653991372106e-05,
+      "loss": 0.14881887435913085,
+      "mean_token_accuracy": 0.9479702442884446,
+      "num_tokens": 8294498.0,
+      "step": 3650
+    },
+    {
+      "epoch": 8.0,
+      "eval_entropy": 0.31352401418345316,
+      "eval_loss": 0.9514623284339905,
+      "eval_mean_token_accuracy": 0.8020065157186418,
+      "eval_num_tokens": 8306000.0,
+      "eval_runtime": 43.9551,
+      "eval_samples_per_second": 22.773,
+      "eval_steps_per_second": 2.867,
+      "step": 3656
+    },
+    {
+      "entropy": 0.1254210752068144,
+      "epoch": 8.096385542168674,
+      "grad_norm": 0.4076255261898041,
+      "learning_rate": 4.394926727987422e-05,
+      "loss": 0.09436151504516602,
+      "mean_token_accuracy": 0.9698121258706758,
+      "num_tokens": 8408606.0,
+      "step": 3700
+    },
+    {
+      "entropy": 0.11122448101639748,
+      "epoch": 8.20591456736035,
+      "grad_norm": 0.42469334602355957,
+      "learning_rate": 3.921033172180787e-05,
+      "loss": 0.0847774600982666,
+      "mean_token_accuracy": 0.972111673951149,
+      "num_tokens": 8523042.0,
+      "step": 3750
+    },
+    {
+      "entropy": 0.1047579725086689,
+      "epoch": 8.315443592552027,
+      "grad_norm": 0.5445597767829895,
+      "learning_rate": 3.47147584676758e-05,
+      "loss": 0.08128767967224121,
+      "mean_token_accuracy": 0.9733622133731842,
+      "num_tokens": 8640478.0,
+      "step": 3800
+    },
+    {
+      "entropy": 0.11103030495345592,
+      "epoch": 8.424972617743702,
+      "grad_norm": 0.4622252881526947,
+      "learning_rate": 3.046910375420573e-05,
+      "loss": 0.08711739540100098,
+      "mean_token_accuracy": 0.9709586310386658,
+      "num_tokens": 8752683.0,
+      "step": 3850
+    },
+    {
+      "entropy": 0.10623676404356956,
+      "epoch": 8.534501642935378,
+      "grad_norm": 0.47963249683380127,
+      "learning_rate": 2.647955934289092e-05,
+      "loss": 0.08331462860107422,
+      "mean_token_accuracy": 0.9721733230352402,
+      "num_tokens": 8869555.0,
+      "step": 3900
+    },
+    {
+      "entropy": 0.1133701204881072,
+      "epoch": 8.644030668127053,
+      "grad_norm": 0.5320031642913818,
+      "learning_rate": 2.275194349007337e-05,
+      "loss": 0.08723929405212402,
+      "mean_token_accuracy": 0.9707023686170578,
+      "num_tokens": 8977869.0,
+      "step": 3950
+    },
+    {
+      "entropy": 0.11115138165652752,
+      "epoch": 8.75355969331873,
+      "grad_norm": 0.46053412556648254,
+      "learning_rate": 1.9291692461737163e-05,
+      "loss": 0.08680099487304688,
+      "mean_token_accuracy": 0.970695189833641,
+      "num_tokens": 9085599.0,
+      "step": 4000
+    },
+    {
+      "entropy": 0.10649645052850247,
+      "epoch": 8.863088718510404,
+      "grad_norm": 0.4637404978275299,
+      "learning_rate": 1.6103852605388375e-05,
+      "loss": 0.08539060592651367,
+      "mean_token_accuracy": 0.9716807705163956,
+      "num_tokens": 9196726.0,
+      "step": 4050
+    },
+    {
+      "entropy": 0.10305101312696933,
+      "epoch": 8.97261774370208,
+      "grad_norm": 0.4751088321208954,
+      "learning_rate": 1.319307299058198e-05,
+      "loss": 0.08058808326721191,
+      "mean_token_accuracy": 0.9734358698129654,
+      "num_tokens": 9314678.0,
+      "step": 4100
+    },
+    {
+      "epoch": 9.0,
+      "eval_entropy": 0.2646038674172901,
+      "eval_loss": 1.105322241783142,
+      "eval_mean_token_accuracy": 0.8011843391827175,
+      "eval_num_tokens": 9344250.0,
+      "eval_runtime": 44.0589,
+      "eval_samples_per_second": 22.72,
+      "eval_steps_per_second": 2.86,
+      "step": 4113
+    },
+    {
+      "entropy": 0.09780798942753763,
+      "epoch": 9.08105147864184,
+      "grad_norm": 0.24395683407783508,
+      "learning_rate": 1.056359862882993e-05,
+      "loss": 0.07237495422363281,
+      "mean_token_accuracy": 0.9756594354456122,
+      "num_tokens": 9425344.0,
+      "step": 4150
+    },
+    {
+      "entropy": 0.08626014020293951,
+      "epoch": 9.190580503833516,
+      "grad_norm": 0.2695149779319763,
+      "learning_rate": 8.219264282777773e-06,
+      "loss": 0.06483334064483642,
+      "mean_token_accuracy": 0.9775187289714813,
+      "num_tokens": 9542855.0,
+      "step": 4200
+    },
+    {
+      "entropy": 0.09083029918372632,
+      "epoch": 9.300109529025193,
+      "grad_norm": 0.2546027898788452,
+      "learning_rate": 6.163488873678151e-06,
+      "loss": 0.06883806705474854,
+      "mean_token_accuracy": 0.9762254297733307,
+      "num_tokens": 9652860.0,
+      "step": 4250
+    },
+    {
+      "entropy": 0.09098202928900719,
+      "epoch": 9.409638554216867,
+      "grad_norm": 0.28322476148605347,
+      "learning_rate": 4.399270495317837e-06,
+      "loss": 0.06978063583374024,
+      "mean_token_accuracy": 0.9756541234254837,
+      "num_tokens": 9764146.0,
+      "step": 4300
+    },
+    {
+      "entropy": 0.0872819459438324,
+      "epoch": 9.519167579408544,
+      "grad_norm": 0.25013017654418945,
+      "learning_rate": 2.9291820416691207e-06,
+      "loss": 0.06817766189575196,
+      "mean_token_accuracy": 0.9770489054918289,
+      "num_tokens": 9877849.0,
+      "step": 4350
+    },
+    {
+      "entropy": 0.08637971702963114,
+      "epoch": 9.628696604600218,
+      "grad_norm": 0.26642662286758423,
+      "learning_rate": 1.7553674546428737e-06,
+      "loss": 0.06635900020599365,
+      "mean_token_accuracy": 0.97729685485363,
+      "num_tokens": 9991781.0,
+      "step": 4400
+    },
+    {
+      "entropy": 0.08398970749229193,
+      "epoch": 9.738225629791895,
+      "grad_norm": 0.20630639791488647,
+      "learning_rate": 8.795385974146928e-07,
+      "loss": 0.06477910995483399,
+      "mean_token_accuracy": 0.9779269409179687,
+      "num_tokens": 10110574.0,
+      "step": 4450
+    },
+    {
+      "entropy": 0.08775710191577674,
+      "epoch": 9.847754654983571,
+      "grad_norm": 0.19058266282081604,
+      "learning_rate": 3.0297275788452486e-07,
+      "loss": 0.0665485143661499,
+      "mean_token_accuracy": 0.9768071568012238,
+      "num_tokens": 10225179.0,
+      "step": 4500
+    },
+    {
+      "entropy": 0.08877926237881184,
+      "epoch": 9.957283680175246,
+      "grad_norm": 0.29807281494140625,
+      "learning_rate": 2.651078591055304e-08,
+      "loss": 0.0695058298110962,
+      "mean_token_accuracy": 0.9764847749471665,
+      "num_tokens": 10336393.0,
+      "step": 4550
+    },
+    {
+      "epoch": 10.0,
+      "eval_entropy": 0.256629646888801,
+      "eval_loss": 1.1647628545761108,
+      "eval_mean_token_accuracy": 0.8009300430615743,
+      "eval_num_tokens": 10382500.0,
+      "eval_runtime": 43.8547,
+      "eval_samples_per_second": 22.825,
+      "eval_steps_per_second": 2.873,
+      "step": 4570
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 4570,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.2535744725559424e+17,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-914/README.md b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-914/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..7ac50e2706e5b4eae8f28e204601afb241c49e55
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-914/README.md
@@ -0,0 +1,209 @@
+---
+base_model: Qwen/Qwen3.5-2B-Base
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:Qwen/Qwen3.5-2B-Base
+- lora
+- sft
+- transformers
+- trl
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.19.1
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-914/adapter_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-914/adapter_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..4d005ec8c649e4abb0922fac729d1444177e0dc1
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-914/adapter_config.json
@@ -0,0 +1,48 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3.5-2B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 256,
+  "lora_bias": false,
+  "lora_dropout": 0.0029143607590526367,
+  "lora_ga_config": null,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.19.1",
+  "qalora_group_size": 16,
+  "r": 128,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "v_proj",
+    "o_proj",
+    "k_proj",
+    "up_proj",
+    "q_proj",
+    "down_proj",
+    "gate_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-914/chat_template.jinja b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-914/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..0ef09f214eaa6d9bca297988afc1454b5827b2c7
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-914/chat_template.jinja
@@ -0,0 +1,154 @@
+{%- set image_count = namespace(value=0) %}
+{%- set video_count = namespace(value=0) %}
+{%- macro render_content(content, do_vision_count, is_system_content=false) %}
+    {%- if content is string %}
+        {{- content }}
+    {%- elif content is iterable and content is not mapping %}
+        {%- for item in content %}
+            {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain images.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set image_count.value = image_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Picture ' ~ image_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|image_pad|><|vision_end|>' }}
+            {%- elif 'video' in item or item.type == 'video' %}
+                {%- if is_system_content %}
+                    {{- raise_exception('System message cannot contain videos.') }}
+                {%- endif %}
+                {%- if do_vision_count %}
+                    {%- set video_count.value = video_count.value + 1 %}
+                {%- endif %}
+                {%- if add_vision_id %}
+                    {{- 'Video ' ~ video_count.value ~ ': ' }}
+                {%- endif %}
+                {{- '<|vision_start|><|video_pad|><|vision_end|>' }}
+            {%- elif 'text' in item %}
+                {{- item.text }}
+            {%- else %}
+                {{- raise_exception('Unexpected item type in content.') }}
+            {%- endif %}
+        {%- endfor %}
+    {%- elif content is none or content is undefined %}
+        {{- '' }}
+    {%- else %}
+        {{- raise_exception('Unexpected content type.') }}
+    {%- endif %}
+{%- endmacro %}
+{%- if not messages %}
+    {{- raise_exception('No messages provided.') }}
+{%- endif %}
+{%- if tools and tools is iterable and tools is not mapping %}
+    {{- '<|im_start|>system\n' }}
+    {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {%- if content %}
+            {{- '\n\n' + content }}
+        {%- endif %}
+    {%- endif %}
+    {{- '<|im_end|>\n' }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {%- set content = render_content(messages[0].content, false, true)|trim %}
+        {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" %}
+        {%- set content = render_content(message.content, false)|trim %}
+        {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
+            {%- set ns.multi_step_tool = false %}
+            {%- set ns.last_query_index = index %}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if ns.multi_step_tool %}
+    {{- raise_exception('No user query found in messages.') }}
+{%- endif %}
+{%- for message in messages %}
+    {%- set content = render_content(message.content, true)|trim %}
+    {%- if message.role == "system" %}
+        {%- if not loop.first %}
+            {{- raise_exception('System message must be at the beginning.') }}
+        {%- endif %}
+    {%- elif message.role == "user" %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- set reasoning_content = reasoning_content|trim %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if tool_call.function is defined %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {%- if loop.first %}
+                    {%- if content|trim %}
+                        {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- else %}
+                        {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                    {%- endif %}
+                {%- else %}
+                    {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
+                {%- endif %}
+                {%- if tool_call.arguments is defined %}
+                    {%- for args_name, args_value in tool_call.arguments|items %}
+                        {{- '<parameter=' + args_name + '>\n' }}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
+                        {{- args_value }}
+                        {{- '\n</parameter>\n' }}
+                    {%- endfor %}
+                {%- endif %}
+                {{- '</function>\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.previtem and loop.previtem.role != "tool" %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if not loop.last and loop.nextitem.role != "tool" %}
+            {{- '<|im_end|>\n' }}
+        {%- elif loop.last %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- else %}
+        {{- raise_exception('Unexpected message role.') }}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is true %}
+        {{- '<think>\n' }}
+    {%- else %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-914/tokenizer_config.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-914/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..b4a37b2a6fd3ab3317cd7bac72855be1a843b2bb
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-914/tokenizer_config.json
@@ -0,0 +1,31 @@
+{
+  "add_prefix_space": false,
+  "audio_bos_token": "<|audio_start|>",
+  "audio_eos_token": "<|audio_end|>",
+  "audio_token": "<|audio_pad|>",
+  "backend": "tokenizers",
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "image_token": "<|image_pad|>",
+  "is_local": false,
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {
+    "audio_bos_token": "<|audio_start|>",
+    "audio_eos_token": "<|audio_end|>",
+    "audio_token": "<|audio_pad|>",
+    "image_token": "<|image_pad|>",
+    "video_token": "<|video_pad|>",
+    "vision_bos_token": "<|vision_start|>",
+    "vision_eos_token": "<|vision_end|>"
+  },
+  "pad_token": "<|endoftext|>",
+  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+  "split_special_tokens": false,
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null,
+  "video_token": "<|video_pad|>",
+  "vision_bos_token": "<|vision_start|>",
+  "vision_eos_token": "<|vision_end|>"
+}
diff --git a/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-914/trainer_state.json b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-914/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..8d3e80147865a7ea0e6cdf2e10b02b8f528cc6b3
--- /dev/null
+++ b/systematicity_original_Estonian/Qwen3.5-2B-Base_systematicity_splits_original_features_train_systematicity_splits_original_features_test2/checkpoint-914/trainer_state.json
@@ -0,0 +1,236 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.0,
+  "eval_steps": 500,
+  "global_step": 914,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "entropy": 1.664013209939003,
+      "epoch": 0.10952902519167579,
+      "grad_norm": 1.825055480003357,
+      "learning_rate": 4.4192733942388133e-05,
+      "loss": 1.5422621154785157,
+      "mean_token_accuracy": 0.6574381196498871,
+      "num_tokens": 113087.0,
+      "step": 50
+    },
+    {
+      "entropy": 0.9298514115810395,
+      "epoch": 0.21905805038335158,
+      "grad_norm": 1.4702502489089966,
+      "learning_rate": 8.928736041421276e-05,
+      "loss": 0.8554936218261718,
+      "mean_token_accuracy": 0.7638577824831009,
+      "num_tokens": 230392.0,
+      "step": 100
+    },
+    {
+      "entropy": 0.8588321554660797,
+      "epoch": 0.32858707557502737,
+      "grad_norm": 1.1502039432525635,
+      "learning_rate": 0.00013438198688603737,
+      "loss": 0.7965679168701172,
+      "mean_token_accuracy": 0.7794932717084885,
+      "num_tokens": 341746.0,
+      "step": 150
+    },
+    {
+      "entropy": 0.8281867235898972,
+      "epoch": 0.43811610076670315,
+      "grad_norm": 1.3328826427459717,
+      "learning_rate": 0.00017947661335786198,
+      "loss": 0.7692495727539063,
+      "mean_token_accuracy": 0.7842007935047149,
+      "num_tokens": 460483.0,
+      "step": 200
+    },
+    {
+      "entropy": 0.8131981420516968,
+      "epoch": 0.547645125958379,
+      "grad_norm": 1.2384045124053955,
+      "learning_rate": 0.00022457123982968662,
+      "loss": 0.7567300415039062,
+      "mean_token_accuracy": 0.7894466006755829,
+      "num_tokens": 570627.0,
+      "step": 250
+    },
+    {
+      "entropy": 0.8158841520547867,
+      "epoch": 0.6571741511500547,
+      "grad_norm": 1.29042387008667,
+      "learning_rate": 0.00026966586630151125,
+      "loss": 0.762672119140625,
+      "mean_token_accuracy": 0.7883876228332519,
+      "num_tokens": 677217.0,
+      "step": 300
+    },
+    {
+      "entropy": 0.8154043215513229,
+      "epoch": 0.7667031763417306,
+      "grad_norm": 1.4341716766357422,
+      "learning_rate": 0.00031476049277333584,
+      "loss": 0.7649243927001953,
+      "mean_token_accuracy": 0.7877721953392028,
+      "num_tokens": 790185.0,
+      "step": 350
+    },
+    {
+      "entropy": 0.8017062950134277,
+      "epoch": 0.8762322015334063,
+      "grad_norm": 1.2653132677078247,
+      "learning_rate": 0.0003598551192451605,
+      "loss": 0.7603777313232422,
+      "mean_token_accuracy": 0.7864453828334809,
+      "num_tokens": 910080.0,
+      "step": 400
+    },
+    {
+      "entropy": 0.8266482776403428,
+      "epoch": 0.9857612267250822,
+      "grad_norm": 1.1987895965576172,
+      "learning_rate": 0.0004049497457169851,
+      "loss": 0.783993911743164,
+      "mean_token_accuracy": 0.7832800775766373,
+      "num_tokens": 1023473.0,
+      "step": 450
+    },
+    {
+      "epoch": 1.0,
+      "eval_entropy": 0.8228033265424153,
+      "eval_loss": 0.8296912908554077,
+      "eval_mean_token_accuracy": 0.769342744634265,
+      "eval_num_tokens": 1038250.0,
+      "eval_runtime": 44.5194,
+      "eval_samples_per_second": 22.485,
+      "eval_steps_per_second": 2.83,
+      "step": 457
+    },
+    {
+      "entropy": 0.8000734815693865,
+      "epoch": 1.0941949616648412,
+      "grad_norm": 1.624888300895691,
+      "learning_rate": 0.00041205884962961745,
+      "loss": 0.7572084045410157,
+      "mean_token_accuracy": 0.7857678087070735,
+      "num_tokens": 1128922.0,
+      "step": 500
+    },
+    {
+      "entropy": 0.7889713329076767,
+      "epoch": 1.203723986856517,
+      "grad_norm": 1.5562400817871094,
+      "learning_rate": 0.0004116562696726163,
+      "loss": 0.757301025390625,
+      "mean_token_accuracy": 0.7859748858213425,
+      "num_tokens": 1245635.0,
+      "step": 550
+    },
+    {
+      "entropy": 0.7770237320661545,
+      "epoch": 1.3132530120481927,
+      "grad_norm": 1.7213724851608276,
+      "learning_rate": 0.00041095388575366294,
+      "loss": 0.7583953857421875,
+      "mean_token_accuracy": 0.7878059667348861,
+      "num_tokens": 1354885.0,
+      "step": 600
+    },
+    {
+      "entropy": 0.7724025696516037,
+      "epoch": 1.4227820372398685,
+      "grad_norm": 1.2281577587127686,
+      "learning_rate": 0.00040995272221269925,
+      "loss": 0.7478806304931641,
+      "mean_token_accuracy": 0.789793593287468,
+      "num_tokens": 1472232.0,
+      "step": 650
+    },
+    {
+      "entropy": 0.7704478764533996,
+      "epoch": 1.5323110624315444,
+      "grad_norm": 1.5951343774795532,
+      "learning_rate": 0.00040865423912273625,
+      "loss": 0.7493467712402344,
+      "mean_token_accuracy": 0.7901222252845764,
+      "num_tokens": 1587494.0,
+      "step": 700
+    },
+    {
+      "entropy": 0.7755234348773956,
+      "epoch": 1.6418400876232202,
+      "grad_norm": 1.2122973203659058,
+      "learning_rate": 0.0004070603301605186,
+      "loss": 0.7549277496337891,
+      "mean_token_accuracy": 0.7899289625883102,
+      "num_tokens": 1696927.0,
+      "step": 750
+    },
+    {
+      "entropy": 0.7561545193195343,
+      "epoch": 1.751369112814896,
+      "grad_norm": 1.262399673461914,
+      "learning_rate": 0.00040517331984483133,
+      "loss": 0.7413782501220703,
+      "mean_token_accuracy": 0.7917980045080185,
+      "num_tokens": 1813415.0,
+      "step": 800
+    },
+    {
+      "entropy": 0.7466552203893662,
+      "epoch": 1.8608981380065717,
+      "grad_norm": 1.2854546308517456,
+      "learning_rate": 0.00040299596014647715,
+      "loss": 0.7443357849121094,
+      "mean_token_accuracy": 0.79294473528862,
+      "num_tokens": 1924911.0,
+      "step": 850
+    },
+    {
+      "entropy": 0.7460890018939972,
+      "epoch": 1.9704271631982475,
+      "grad_norm": 1.0535873174667358,
+      "learning_rate": 0.00040053142647486815,
+      "loss": 0.742437973022461,
+      "mean_token_accuracy": 0.7911104756593704,
+      "num_tokens": 2044650.0,
+      "step": 900
+    },
+    {
+      "epoch": 2.0,
+      "eval_entropy": 0.7546336958332668,
+      "eval_loss": 0.7908218502998352,
+      "eval_mean_token_accuracy": 0.7816079741432553,
+      "eval_num_tokens": 2076500.0,
+      "eval_runtime": 43.6577,
+      "eval_samples_per_second": 22.928,
+      "eval_steps_per_second": 2.886,
+      "step": 914
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 4570,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 10,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 4.486683349186176e+16,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}