Upload 18 files

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,3 +1,58 @@
----
-license: mit
----
+---
+library_name: peft
+license: other
+base_model: unsloth/gemma-3-1b-it
+tags:
+- llama-factory
+- lora
+- generated_from_trainer
+model-index:
+- name: train_2025-07-14-13-26-41
+  results: []
+---
+
+<!-- This model card has been generated automatically according to the information the Trainer had access to. You
+should probably proofread and complete it, then remove this comment. -->
+
+# train_2025-07-14-13-26-41
+
+This model is a fine-tuned version of [unsloth/gemma-3-1b-it](https://huggingface.co/unsloth/gemma-3-1b-it) on the oit dataset.
+
+## Model description
+
+More information needed
+
+## Intended uses & limitations
+
+More information needed
+
+## Training and evaluation data
+
+More information needed
+
+## Training procedure
+
+### Training hyperparameters
+
+The following hyperparameters were used during training:
+- learning_rate: 5e-05
+- train_batch_size: 3
+- eval_batch_size: 8
+- seed: 42
+- gradient_accumulation_steps: 8
+- total_train_batch_size: 24
+- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
+- lr_scheduler_type: cosine
+- num_epochs: 3.0
+
+### Training results
+
+
+
+### Framework versions
+
+- PEFT 0.15.2
+- Transformers 4.52.4
+- Pytorch 2.7.1+cu128
+- Datasets 3.6.0
+- Tokenizers 0.21.1

adapter_config.json

@@ -0,0 +1,39 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "unsloth/gemma-3-1b-it",
+  "bias": "none",
+  "corda_config": null,
+  "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": 16,
+  "lora_bias": false,
+  "lora_dropout": 0,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "r": 8,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "q_proj",
+    "down_proj",
+    "up_proj",
+    "k_proj",
+    "gate_proj",
+    "v_proj",
+    "o_proj"
+  ],
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_rslora": false
+}

adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:515a17c1f2cbe015ca7b84fee3e8d21123b1f97de1f128c50530d5e0472e7104
+size 26139264

added_tokens.json

@@ -0,0 +1,3 @@
+{
+  "<image_soft_token>": 262144
+}

all_results.json

@@ -0,0 +1,9 @@
+{
+    "epoch": 3.0,
+    "num_input_tokens_seen": 85968,
+    "total_flos": 363344926666752.0,
+    "train_loss": 3.3720982670783997,
+    "train_runtime": 88.7993,
+    "train_samples_per_second": 9.155,
+    "train_steps_per_second": 0.405
+}

chat_template.jinja

@@ -0,0 +1,47 @@
+{{ bos_token }}
+{%- if messages[0]['role'] == 'system' -%}
+    {%- if messages[0]['content'] is string -%}
+        {%- set first_user_prefix = messages[0]['content'] + '
+
+' -%}
+    {%- else -%}
+        {%- set first_user_prefix = messages[0]['content'][0]['text'] + '
+
+' -%}
+    {%- endif -%}
+    {%- set loop_messages = messages[1:] -%}
+{%- else -%}
+    {%- set first_user_prefix = "" -%}
+    {%- set loop_messages = messages -%}
+{%- endif -%}
+{%- for message in loop_messages -%}
+    {%- if (message['role'] == 'user') != (loop.index0 % 2 == 0) -%}
+        {{ raise_exception("Conversation roles must alternate user/assistant/user/assistant/...") }}
+    {%- endif -%}
+    {%- if (message['role'] == 'assistant') -%}
+        {%- set role = "model" -%}
+    {%- else -%}
+        {%- set role = message['role'] -%}
+    {%- endif -%}
+    {{ '<start_of_turn>' + role + '
+' + (first_user_prefix if loop.first else "") }}
+    {%- if message['content'] is string -%}
+        {{ message['content'] | trim }}
+    {%- elif message['content'] is iterable -%}
+        {%- for item in message['content'] -%}
+            {%- if item['type'] == 'image' -%}
+                {{ '<start_of_image>' }}
+            {%- elif item['type'] == 'text' -%}
+                {{ item['text'] | trim }}
+            {%- endif -%}
+        {%- endfor -%}
+    {%- else -%}
+        {{ raise_exception("Invalid content type") }}
+    {%- endif -%}
+    {{ '<end_of_turn>
+' }}
+{%- endfor -%}
+{%- if add_generation_prompt -%}
+    {{'<start_of_turn>model
+'}}
+{%- endif -%}

llamaboard_config.yaml

@@ -0,0 +1,85 @@
+top.booster: auto
+top.checkpoint_path: []
+top.finetuning_type: lora
+top.model_name: Gemma-3-1B-Instruct
+top.quantization_bit: '4'
+top.quantization_method: bnb
+top.rope_scaling: none
+top.template: gemma
+train.additional_target: ''
+train.apollo_rank: 16
+train.apollo_scale: 32
+train.apollo_target: all
+train.apollo_update_interval: 200
+train.badam_mode: layer
+train.badam_switch_interval: 50
+train.badam_switch_mode: ascending
+train.badam_update_ratio: 0.05
+train.batch_size: 3
+train.compute_type: bf16
+train.create_new_adapter: false
+train.cutoff_len: 4096
+train.dataset:
+- oit
+train.dataset_dir: data
+train.ds_offload: false
+train.ds_stage: none
+train.enable_thinking: true
+train.extra_args: '{"optim": "adamw_torch"}'
+train.freeze_extra_modules: ''
+train.freeze_language_model: false
+train.freeze_multi_modal_projector: true
+train.freeze_trainable_layers: 2
+train.freeze_trainable_modules: all
+train.freeze_vision_tower: true
+train.galore_rank: 16
+train.galore_scale: 2
+train.galore_target: all
+train.galore_update_interval: 200
+train.gradient_accumulation_steps: 8
+train.image_max_pixels: 768*768
+train.image_min_pixels: 32*32
+train.learning_rate: 5e-5
+train.logging_steps: 5
+train.lora_alpha: 16
+train.lora_dropout: 0
+train.lora_rank: 8
+train.lora_target: ''
+train.loraplus_lr_ratio: 0
+train.lr_scheduler_type: cosine
+train.mask_history: false
+train.max_grad_norm: '1.0'
+train.max_samples: '100000'
+train.neat_packing: false
+train.neftune_alpha: 0
+train.num_train_epochs: '3.0'
+train.packing: false
+train.ppo_score_norm: false
+train.ppo_whiten_rewards: false
+train.pref_beta: 0.1
+train.pref_ftx: 0
+train.pref_loss: sigmoid
+train.report_to: none
+train.resize_vocab: false
+train.reward_model: []
+train.save_steps: 100
+train.swanlab_api_key: ''
+train.swanlab_link: ''
+train.swanlab_mode: cloud
+train.swanlab_project: llamafactory
+train.swanlab_run_name: ''
+train.swanlab_workspace: ''
+train.train_on_prompt: false
+train.training_stage: Supervised Fine-Tuning
+train.use_apollo: false
+train.use_badam: false
+train.use_dora: false
+train.use_galore: false
+train.use_llama_pro: false
+train.use_pissa: false
+train.use_rslora: false
+train.use_swanlab: false
+train.val_size: 0
+train.video_max_pixels: 256*256
+train.video_min_pixels: 16*16
+train.warmup_steps: 0

running_log.txt

@@ -0,0 +1,197 @@
+[INFO|2025-07-14 13:33:21] tokenization_utils_base.py:2023 >> loading file tokenizer.model from cache at /home/ats-ai/.cache/huggingface/hub/models--unsloth--gemma-3-1b-it/snapshots/5b11413a10db4e486ef16a20101fd028f8f2499c/tokenizer.model
+[INFO|2025-07-14 13:33:21] tokenization_utils_base.py:2023 >> loading file tokenizer.json from cache at /home/ats-ai/.cache/huggingface/hub/models--unsloth--gemma-3-1b-it/snapshots/5b11413a10db4e486ef16a20101fd028f8f2499c/tokenizer.json
+[INFO|2025-07-14 13:33:21] tokenization_utils_base.py:2023 >> loading file added_tokens.json from cache at /home/ats-ai/.cache/huggingface/hub/models--unsloth--gemma-3-1b-it/snapshots/5b11413a10db4e486ef16a20101fd028f8f2499c/added_tokens.json
+[INFO|2025-07-14 13:33:21] tokenization_utils_base.py:2023 >> loading file special_tokens_map.json from cache at /home/ats-ai/.cache/huggingface/hub/models--unsloth--gemma-3-1b-it/snapshots/5b11413a10db4e486ef16a20101fd028f8f2499c/special_tokens_map.json
+[INFO|2025-07-14 13:33:21] tokenization_utils_base.py:2023 >> loading file tokenizer_config.json from cache at /home/ats-ai/.cache/huggingface/hub/models--unsloth--gemma-3-1b-it/snapshots/5b11413a10db4e486ef16a20101fd028f8f2499c/tokenizer_config.json
+[INFO|2025-07-14 13:33:21] tokenization_utils_base.py:2023 >> loading file chat_template.jinja from cache at /home/ats-ai/.cache/huggingface/hub/models--unsloth--gemma-3-1b-it/snapshots/5b11413a10db4e486ef16a20101fd028f8f2499c/chat_template.jinja
+[INFO|2025-07-14 13:33:23] image_processing_base.py:380 >> loading configuration file config.json from cache at /home/ats-ai/.cache/huggingface/hub/models--unsloth--gemma-3-1b-it/snapshots/5b11413a10db4e486ef16a20101fd028f8f2499c/config.json
+[INFO|2025-07-14 13:33:23] logging.py:143 >> Failed to load processor: unsloth/gemma-3-1b-it does not appear to have a file named preprocessor_config.json. Checkout 'https://huggingface.co/unsloth/gemma-3-1b-it/tree/main' for available files..
+[INFO|2025-07-14 13:33:23] logging.py:143 >> Loading dataset oit.json...
+[INFO|2025-07-14 13:33:30] configuration_utils.py:698 >> loading configuration file config.json from cache at /home/ats-ai/.cache/huggingface/hub/models--unsloth--gemma-3-1b-it/snapshots/5b11413a10db4e486ef16a20101fd028f8f2499c/config.json
+[INFO|2025-07-14 13:33:30] configuration_utils.py:770 >> Model config Gemma3TextConfig {
+  "architectures": [
+    "Gemma3ForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "attn_logit_softcapping": null,
+  "bos_token_id": 2,
+  "cache_implementation": "hybrid",
+  "eos_token_id": 106,
+  "final_logit_softcapping": null,
+  "head_dim": 256,
+  "hidden_activation": "gelu_pytorch_tanh",
+  "hidden_size": 1152,
+  "initializer_range": 0.02,
+  "intermediate_size": 6912,
+  "max_position_embeddings": 32768,
+  "model_type": "gemma3_text",
+  "num_attention_heads": 4,
+  "num_hidden_layers": 26,
+  "num_key_value_heads": 1,
+  "pad_token_id": 0,
+  "query_pre_attn_scalar": 256,
+  "rms_norm_eps": 1e-06,
+  "rope_local_base_freq": 10000,
+  "rope_scaling": null,
+  "rope_theta": 1000000,
+  "sliding_window": 512,
+  "sliding_window_pattern": 6,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.52.4",
+  "unsloth_fixed": true,
+  "use_cache": true,
+  "vocab_size": 262144
+}
+
+[INFO|2025-07-14 13:33:30] logging.py:143 >> Quantizing model to 4 bit with bitsandbytes.
+[INFO|2025-07-14 13:33:30] logging.py:143 >> KV cache is disabled during training.
+[INFO|2025-07-14 13:33:47] modeling_utils.py:1151 >> loading weights file model.safetensors from cache at /home/ats-ai/.cache/huggingface/hub/models--unsloth--gemma-3-1b-it/snapshots/5b11413a10db4e486ef16a20101fd028f8f2499c/model.safetensors
+[INFO|2025-07-14 13:33:47] modeling_utils.py:2241 >> Instantiating Gemma3ForCausalLM model under default dtype torch.bfloat16.
+[WARNING|2025-07-14 13:33:47] configuration_utils.py:839 >> The following generation flags are not valid and may be ignored: ['cache_implementation'].
+[INFO|2025-07-14 13:33:47] configuration_utils.py:840 >> - `cache_implementation`: You have set `use_cache` to `False`, but cache_implementation is set to hybrid. cache_implementation will have no effect.
+If you're using a pretrained model, note that some of these attributes may be set through the model's `generation_config.json` file.
+[WARNING|2025-07-14 13:33:47] configuration_utils.py:839 >> The following generation flags are not valid and may be ignored: ['cache_implementation'].
+[INFO|2025-07-14 13:33:47] configuration_utils.py:840 >> - `cache_implementation`: You have set `use_cache` to `False`, but cache_implementation is set to hybrid. cache_implementation will have no effect.
+If you're using a pretrained model, note that some of these attributes may be set through the model's `generation_config.json` file.
+[INFO|2025-07-14 13:33:47] configuration_utils.py:1135 >> Generate config GenerationConfig {
+  "bos_token_id": 2,
+  "cache_implementation": "hybrid",
+  "eos_token_id": 106,
+  "pad_token_id": 0,
+  "use_cache": false
+}
+
+[INFO|2025-07-14 13:33:48] modeling_utils.py:5131 >> All model checkpoint weights were used when initializing Gemma3ForCausalLM.
+
+[INFO|2025-07-14 13:33:48] modeling_utils.py:5139 >> All the weights of Gemma3ForCausalLM were initialized from the model checkpoint at unsloth/gemma-3-1b-it.
+If your task is similar to the task the model of the checkpoint was trained on, you can already use Gemma3ForCausalLM for predictions without further training.
+[INFO|2025-07-14 13:33:48] configuration_utils.py:1090 >> loading configuration file generation_config.json from cache at /home/ats-ai/.cache/huggingface/hub/models--unsloth--gemma-3-1b-it/snapshots/5b11413a10db4e486ef16a20101fd028f8f2499c/generation_config.json
+[INFO|2025-07-14 13:33:48] configuration_utils.py:1135 >> Generate config GenerationConfig {
+  "bos_token_id": 2,
+  "cache_implementation": "hybrid",
+  "do_sample": true,
+  "eos_token_id": [
+    1,
+    106
+  ],
+  "max_length": 32768,
+  "pad_token_id": 0,
+  "top_k": 64,
+  "top_p": 0.95
+}
+
+[INFO|2025-07-14 13:33:48] logging.py:143 >> Gradient checkpointing enabled.
+[INFO|2025-07-14 13:33:48] logging.py:143 >> Using torch SDPA for faster training and inference.
+[INFO|2025-07-14 13:33:48] logging.py:143 >> Upcasting trainable params to float32.
+[INFO|2025-07-14 13:33:48] logging.py:143 >> Fine-tuning method: LoRA
+[INFO|2025-07-14 13:33:48] logging.py:143 >> Found linear modules: q_proj,down_proj,up_proj,k_proj,gate_proj,v_proj,o_proj
+[INFO|2025-07-14 13:33:48] logging.py:143 >> trainable params: 6,522,880 || all params: 1,006,408,832 || trainable%: 0.6481
+[INFO|2025-07-14 13:33:48] trainer.py:756 >> Using auto half precision backend
+[INFO|2025-07-14 13:33:48] trainer.py:2409 >> ***** Running training *****
+[INFO|2025-07-14 13:33:48] trainer.py:2410 >>   Num examples = 271
+[INFO|2025-07-14 13:33:48] trainer.py:2411 >>   Num Epochs = 3
+[INFO|2025-07-14 13:33:48] trainer.py:2412 >>   Instantaneous batch size per device = 3
+[INFO|2025-07-14 13:33:48] trainer.py:2415 >>   Total train batch size (w. parallel, distributed & accumulation) = 24
+[INFO|2025-07-14 13:33:48] trainer.py:2416 >>   Gradient Accumulation steps = 8
+[INFO|2025-07-14 13:33:48] trainer.py:2417 >>   Total optimization steps = 36
+[INFO|2025-07-14 13:33:48] trainer.py:2418 >>   Number of trainable parameters = 6,522,880
+[WARNING|2025-07-14 13:33:48] logging.py:328 >> It is strongly recommended to train Gemma3 models with the `eager` attention implementation instead of `sdpa`. Use `eager` with `AutoModelForCausalLM.from_pretrained('<path-to-checkpoint>', attn_implementation='eager')`.
+[INFO|2025-07-14 13:34:01] logging.py:143 >> {'loss': 5.4762, 'learning_rate': 4.8492e-05, 'epoch': 0.44, 'throughput': 1034.76}
+[INFO|2025-07-14 13:34:13] logging.py:143 >> {'loss': 3.9653, 'learning_rate': 4.2678e-05, 'epoch': 0.88, 'throughput': 1031.28}
+[INFO|2025-07-14 13:34:24] logging.py:143 >> {'loss': 3.2543, 'learning_rate': 3.3551e-05, 'epoch': 1.26, 'throughput': 1027.32}
+[INFO|2025-07-14 13:34:38] logging.py:143 >> {'loss': 2.9243, 'learning_rate': 2.2821e-05, 'epoch': 1.70, 'throughput': 1000.40}
+[INFO|2025-07-14 13:34:49] logging.py:143 >> {'loss': 2.6278, 'learning_rate': 1.2500e-05, 'epoch': 2.09, 'throughput': 1001.62}
+[INFO|2025-07-14 13:35:03] logging.py:143 >> {'loss': 2.6988, 'learning_rate': 4.5212e-06, 'epoch': 2.53, 'throughput': 981.11}
+[INFO|2025-07-14 13:35:16] logging.py:143 >> {'loss': 2.7677, 'learning_rate': 3.7981e-07, 'epoch': 2.97, 'throughput': 977.06}
+[INFO|2025-07-14 13:35:17] trainer.py:3993 >> Saving model checkpoint to saves/Gemma-3-1B-Instruct/lora/train_2025-07-14-13-26-41/checkpoint-36
+[INFO|2025-07-14 13:35:17] configuration_utils.py:698 >> loading configuration file config.json from cache at /home/ats-ai/.cache/huggingface/hub/models--unsloth--gemma-3-1b-it/snapshots/5b11413a10db4e486ef16a20101fd028f8f2499c/config.json
+[INFO|2025-07-14 13:35:17] configuration_utils.py:770 >> Model config Gemma3TextConfig {
+  "architectures": [
+    "Gemma3ForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "attn_logit_softcapping": null,
+  "bos_token_id": 2,
+  "cache_implementation": "hybrid",
+  "eos_token_id": 106,
+  "final_logit_softcapping": null,
+  "head_dim": 256,
+  "hidden_activation": "gelu_pytorch_tanh",
+  "hidden_size": 1152,
+  "initializer_range": 0.02,
+  "intermediate_size": 6912,
+  "max_position_embeddings": 32768,
+  "model_type": "gemma3_text",
+  "num_attention_heads": 4,
+  "num_hidden_layers": 26,
+  "num_key_value_heads": 1,
+  "pad_token_id": 0,
+  "query_pre_attn_scalar": 256,
+  "rms_norm_eps": 1e-06,
+  "rope_local_base_freq": 10000,
+  "rope_scaling": null,
+  "rope_theta": 1000000,
+  "sliding_window": 512,
+  "sliding_window_pattern": 6,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.52.4",
+  "unsloth_fixed": true,
+  "use_cache": true,
+  "vocab_size": 262144
+}
+
+[INFO|2025-07-14 13:35:17] tokenization_utils_base.py:2356 >> chat template saved in saves/Gemma-3-1B-Instruct/lora/train_2025-07-14-13-26-41/checkpoint-36/chat_template.jinja
+[INFO|2025-07-14 13:35:17] tokenization_utils_base.py:2525 >> tokenizer config file saved in saves/Gemma-3-1B-Instruct/lora/train_2025-07-14-13-26-41/checkpoint-36/tokenizer_config.json
+[INFO|2025-07-14 13:35:17] tokenization_utils_base.py:2534 >> Special tokens file saved in saves/Gemma-3-1B-Instruct/lora/train_2025-07-14-13-26-41/checkpoint-36/special_tokens_map.json
+[INFO|2025-07-14 13:35:17] trainer.py:2676 >> 
+
+Training completed. Do not forget to share your model on huggingface.co/models =)
+
+
+[INFO|2025-07-14 13:35:17] trainer.py:3993 >> Saving model checkpoint to saves/Gemma-3-1B-Instruct/lora/train_2025-07-14-13-26-41
+[INFO|2025-07-14 13:35:17] configuration_utils.py:698 >> loading configuration file config.json from cache at /home/ats-ai/.cache/huggingface/hub/models--unsloth--gemma-3-1b-it/snapshots/5b11413a10db4e486ef16a20101fd028f8f2499c/config.json
+[INFO|2025-07-14 13:35:17] configuration_utils.py:770 >> Model config Gemma3TextConfig {
+  "architectures": [
+    "Gemma3ForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "attn_logit_softcapping": null,
+  "bos_token_id": 2,
+  "cache_implementation": "hybrid",
+  "eos_token_id": 106,
+  "final_logit_softcapping": null,
+  "head_dim": 256,
+  "hidden_activation": "gelu_pytorch_tanh",
+  "hidden_size": 1152,
+  "initializer_range": 0.02,
+  "intermediate_size": 6912,
+  "max_position_embeddings": 32768,
+  "model_type": "gemma3_text",
+  "num_attention_heads": 4,
+  "num_hidden_layers": 26,
+  "num_key_value_heads": 1,
+  "pad_token_id": 0,
+  "query_pre_attn_scalar": 256,
+  "rms_norm_eps": 1e-06,
+  "rope_local_base_freq": 10000,
+  "rope_scaling": null,
+  "rope_theta": 1000000,
+  "sliding_window": 512,
+  "sliding_window_pattern": 6,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.52.4",
+  "unsloth_fixed": true,
+  "use_cache": true,
+  "vocab_size": 262144
+}
+
+[INFO|2025-07-14 13:35:17] tokenization_utils_base.py:2356 >> chat template saved in saves/Gemma-3-1B-Instruct/lora/train_2025-07-14-13-26-41/chat_template.jinja
+[INFO|2025-07-14 13:35:17] tokenization_utils_base.py:2525 >> tokenizer config file saved in saves/Gemma-3-1B-Instruct/lora/train_2025-07-14-13-26-41/tokenizer_config.json
+[INFO|2025-07-14 13:35:17] tokenization_utils_base.py:2534 >> Special tokens file saved in saves/Gemma-3-1B-Instruct/lora/train_2025-07-14-13-26-41/special_tokens_map.json
+[WARNING|2025-07-14 13:35:18] logging.py:148 >> No metric eval_loss to plot.
+[WARNING|2025-07-14 13:35:18] logging.py:148 >> No metric eval_accuracy to plot.
+[INFO|2025-07-14 13:35:18] modelcard.py:450 >> Dropping the following result as it does not have all the necessary fields:
+{'task': {'name': 'Causal Language Modeling', 'type': 'text-generation'}}

special_tokens_map.json

@@ -0,0 +1,33 @@
+{
+  "boi_token": "<start_of_image>",
+  "bos_token": {
+    "content": "<bos>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eoi_token": "<end_of_image>",
+  "eos_token": {
+    "content": "<end_of_turn>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "image_token": "<image_soft_token>",
+  "pad_token": {
+    "content": "<pad>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4667f2089529e8e7657cfb6d1c19910ae71ff5f28aa7ab2ff2763330affad795
+size 33384568

tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1299c11d7cf632ef3b4e11937501358ada021bbdf7c47638d13c0ee982f2e79c
+size 4689074

train_results.json

@@ -0,0 +1,9 @@
+{
+    "epoch": 3.0,
+    "num_input_tokens_seen": 85968,
+    "total_flos": 363344926666752.0,
+    "train_loss": 3.3720982670783997,
+    "train_runtime": 88.7993,
+    "train_samples_per_second": 9.155,
+    "train_steps_per_second": 0.405
+}

trainer_log.jsonl

@@ -0,0 +1,8 @@
+{"current_steps": 5, "total_steps": 36, "loss": 5.4762, "lr": 4.849231551964771e-05, "epoch": 0.43956043956043955, "percentage": 13.89, "elapsed_time": "0:00:12", "remaining_time": "0:01:19", "throughput": 1034.76, "total_tokens": 13272}
+{"current_steps": 10, "total_steps": 36, "loss": 3.9653, "lr": 4.267766952966369e-05, "epoch": 0.8791208791208791, "percentage": 27.78, "elapsed_time": "0:00:24", "remaining_time": "0:01:03", "throughput": 1031.28, "total_tokens": 25368}
+{"current_steps": 15, "total_steps": 36, "loss": 3.2543, "lr": 3.355050358314172e-05, "epoch": 1.2637362637362637, "percentage": 41.67, "elapsed_time": "0:00:35", "remaining_time": "0:00:49", "throughput": 1027.32, "total_tokens": 36568}
+{"current_steps": 20, "total_steps": 36, "loss": 2.9243, "lr": 2.2821106431308544e-05, "epoch": 1.7032967032967035, "percentage": 55.56, "elapsed_time": "0:00:49", "remaining_time": "0:00:39", "throughput": 1000.4, "total_tokens": 49360}
+{"current_steps": 25, "total_steps": 36, "loss": 2.6278, "lr": 1.2500000000000006e-05, "epoch": 2.087912087912088, "percentage": 69.44, "elapsed_time": "0:01:00", "remaining_time": "0:00:26", "throughput": 1001.62, "total_tokens": 60992}
+{"current_steps": 30, "total_steps": 36, "loss": 2.6988, "lr": 4.521198892775203e-06, "epoch": 2.5274725274725274, "percentage": 83.33, "elapsed_time": "0:01:14", "remaining_time": "0:00:14", "throughput": 981.11, "total_tokens": 72872}
+{"current_steps": 35, "total_steps": 36, "loss": 2.7677, "lr": 3.7980617469479953e-07, "epoch": 2.967032967032967, "percentage": 97.22, "elapsed_time": "0:01:27", "remaining_time": "0:00:02", "throughput": 977.06, "total_tokens": 85304}
+{"current_steps": 36, "total_steps": 36, "epoch": 3.0, "percentage": 100.0, "elapsed_time": "0:01:28", "remaining_time": "0:00:00", "throughput": 968.13, "total_tokens": 85968}

trainer_state.json

@@ -0,0 +1,114 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.0,
+  "eval_steps": 500,
+  "global_step": 36,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.43956043956043955,
+      "grad_norm": 9.493675231933594,
+      "learning_rate": 4.849231551964771e-05,
+      "loss": 5.4762,
+      "num_input_tokens_seen": 13272,
+      "step": 5,
+      "train_runtime": 12.8278,
+      "train_tokens_per_second": 1034.626
+    },
+    {
+      "epoch": 0.8791208791208791,
+      "grad_norm": 4.911149024963379,
+      "learning_rate": 4.267766952966369e-05,
+      "loss": 3.9653,
+      "num_input_tokens_seen": 25368,
+      "step": 10,
+      "train_runtime": 24.6003,
+      "train_tokens_per_second": 1031.207
+    },
+    {
+      "epoch": 1.2637362637362637,
+      "grad_norm": 3.035015821456909,
+      "learning_rate": 3.355050358314172e-05,
+      "loss": 3.2543,
+      "num_input_tokens_seen": 36568,
+      "step": 15,
+      "train_runtime": 35.5974,
+      "train_tokens_per_second": 1027.266
+    },
+    {
+      "epoch": 1.7032967032967035,
+      "grad_norm": 2.246847152709961,
+      "learning_rate": 2.2821106431308544e-05,
+      "loss": 2.9243,
+      "num_input_tokens_seen": 49360,
+      "step": 20,
+      "train_runtime": 49.3421,
+      "train_tokens_per_second": 1000.362
+    },
+    {
+      "epoch": 2.087912087912088,
+      "grad_norm": 1.6257575750350952,
+      "learning_rate": 1.2500000000000006e-05,
+      "loss": 2.6278,
+      "num_input_tokens_seen": 60992,
+      "step": 25,
+      "train_runtime": 60.8949,
+      "train_tokens_per_second": 1001.594
+    },
+    {
+      "epoch": 2.5274725274725274,
+      "grad_norm": 1.944761872291565,
+      "learning_rate": 4.521198892775203e-06,
+      "loss": 2.6988,
+      "num_input_tokens_seen": 72872,
+      "step": 30,
+      "train_runtime": 74.2765,
+      "train_tokens_per_second": 981.091
+    },
+    {
+      "epoch": 2.967032967032967,
+      "grad_norm": 1.752797245979309,
+      "learning_rate": 3.7980617469479953e-07,
+      "loss": 2.7677,
+      "num_input_tokens_seen": 85304,
+      "step": 35,
+      "train_runtime": 87.3084,
+      "train_tokens_per_second": 977.042
+    },
+    {
+      "epoch": 3.0,
+      "num_input_tokens_seen": 85968,
+      "step": 36,
+      "total_flos": 363344926666752.0,
+      "train_loss": 3.3720982670783997,
+      "train_runtime": 88.7993,
+      "train_samples_per_second": 9.155,
+      "train_steps_per_second": 0.405
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 36,
+  "num_input_tokens_seen": 85968,
+  "num_train_epochs": 3,
+  "save_steps": 100,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 363344926666752.0,
+  "train_batch_size": 3,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:017320e2e7dc050c148ab1218dc90c9ede1f72f067caf9b74b319c767b40edeb
+size 6161

training_args.yaml

@@ -0,0 +1,37 @@
+bf16: true
+cutoff_len: 4096
+dataset: oit
+dataset_dir: data
+ddp_timeout: 180000000
+do_train: true
+double_quantization: true
+enable_thinking: true
+finetuning_type: lora
+flash_attn: auto
+gradient_accumulation_steps: 8
+include_num_input_tokens_seen: true
+learning_rate: 5.0e-05
+logging_steps: 5
+lora_alpha: 16
+lora_dropout: 0
+lora_rank: 8
+lora_target: all
+lr_scheduler_type: cosine
+max_grad_norm: 1.0
+max_samples: 100000
+model_name_or_path: unsloth/gemma-3-1b-it
+num_train_epochs: 3.0
+optim: adamw_torch
+output_dir: saves/Gemma-3-1B-Instruct/lora/train_2025-07-14-13-26-41
+packing: false
+per_device_train_batch_size: 3
+plot_loss: true
+preprocessing_num_workers: 16
+quantization_bit: 4
+quantization_method: bnb
+report_to: none
+save_steps: 100
+stage: sft
+template: gemma
+trust_remote_code: true
+warmup_steps: 0