Full backup of working directory

.gitattributes

@@ -33,3 +33,11 @@ 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
+data/merged.jsonl filter=lfs diff=lfs merge=lfs -text
+gemma4_e2b_out/checkpoint-1000/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+gemma4_gguf/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+lora_weights_final/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+unsloth_compiled_cache/__pycache__/UnslothDPOTrainer.cpython-312.pyc filter=lfs diff=lfs merge=lfs -text
+unsloth_compiled_cache/__pycache__/UnslothGRPOTrainer.cpython-312.pyc filter=lfs diff=lfs merge=lfs -text
+unsloth_compiled_cache/__pycache__/UnslothKTOTrainer.cpython-312.pyc filter=lfs diff=lfs merge=lfs -text
+unsloth_compiled_cache/__pycache__/UnslothRLOOTrainer.cpython-312.pyc filter=lfs diff=lfs merge=lfs -text

data/merged.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8c41510b2b0f6b95e6ae376da19a095eb2d034c1455b2e3d609d9a8f194e3c27
+size 1141102581

gemma4_e2b_out/README.md

@@ -0,0 +1,59 @@
+---
+base_model: unsloth/gemma-4-e2b-it-unsloth-bnb-4bit
+library_name: transformers
+model_name: gemma4_e2b_out
+tags:
+- generated_from_trainer
+- trl
+- unsloth
+- sft
+licence: license
+---
+
+# Model Card for gemma4_e2b_out
+
+This model is a fine-tuned version of [unsloth/gemma-4-e2b-it-unsloth-bnb-4bit](https://huggingface.co/unsloth/gemma-4-e2b-it-unsloth-bnb-4bit).
+It has been trained using [TRL](https://github.com/huggingface/trl).
+
+## Quick start
+
+```python
+from transformers import pipeline
+
+question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
+generator = pipeline("text-generation", model="None", device="cuda")
+output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
+print(output["generated_text"])
+```
+
+## Training procedure
+
+ 
+
+
+
+This model was trained with SFT.
+
+### Framework versions
+
+- TRL: 1.7.0
+- Transformers: 5.5.0
+- Pytorch: 2.10.0+cu128
+- Datasets: 5.0.0
+- Tokenizers: 0.22.2
+
+## Citations
+
+
+
+Cite TRL as:
+    
+```bibtex
+@software{vonwerra2020trl,
+  title   = {{TRL: Transformers Reinforcement Learning}},
+  author  = {von Werra, Leandro and Belkada, Younes and Tunstall, Lewis and Beeching, Edward and Thrush, Tristan and Lambert, Nathan and Huang, Shengyi and Rasul, Kashif and Gallouédec, Quentin},
+  license = {Apache-2.0},
+  url     = {https://github.com/huggingface/trl},
+  year    = {2020}
+}
+```

gemma4_e2b_out/checkpoint-1000/README.md

@@ -0,0 +1,210 @@
+---
+base_model: unsloth/gemma-4-e2b-it-unsloth-bnb-4bit
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:unsloth/gemma-4-e2b-it-unsloth-bnb-4bit
+- lora
+- sft
+- transformers
+- trl
+- unsloth
+---
+
+# 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

gemma4_e2b_out/checkpoint-1000/adapter_config.json

@@ -0,0 +1,44 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Gemma4ForConditionalGeneration",
+    "parent_library": "transformers.models.gemma4.modeling_gemma4",
+    "unsloth_fixed": true
+  },
+  "base_model_name_or_path": "unsloth/gemma-4-e2b-it-unsloth-bnb-4bit",
+  "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,
+  "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": 32,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": "(?:.*?(?:language|text).*?(?:self_attn|attention|attn|mixer|mlp|feed_forward|ffn|dense|mixer).*?(?:q_proj|k_proj|v_proj|o_proj|gate_proj|up_proj|down_proj))|(?:\\bmodel\\.layers\\.[\\d]{1,}\\.(?:self_attn|attention|attn|mixer|mlp|feed_forward|ffn|dense|mixer)\\.(?:(?:q_proj|k_proj|v_proj|o_proj|gate_proj|up_proj|down_proj)))",
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

gemma4_e2b_out/checkpoint-1000/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f14e7d5c4c6cfe03e5f99ffea455059061c7edc6b443e909b0c236b0bc8913cb
+size 101424416

gemma4_e2b_out/checkpoint-1000/chat_template.jinja

@@ -0,0 +1,70 @@
+{{ bos_token }}{%- macro strip_thinking(text) -%}
+    {%- set ns = namespace(result='') -%}
+    {%- for part in text.split('<channel|>') -%}
+        {%- if '<|channel>' in part -%}
+            {%- set ns.result = ns.result + part.split('<|channel>')[0] -%}
+        {%- else -%}
+            {%- set ns.result = ns.result + part -%}
+        {%- endif -%}
+    {%- endfor -%}
+    {{- ns.result | trim -}}
+{%- endmacro -%}
+{%- set thinking = enable_thinking is defined and enable_thinking -%}
+{%- set loop_messages = messages -%}
+{%- if messages[0]['role'] in ['system', 'developer'] or thinking -%}
+    {{ '<|turn>system
+' }}
+    {%- if thinking -%}
+        {{ '<|think|>
+' }}
+    {%- endif -%}
+    {%- if messages[0]['role'] in ['system', 'developer'] -%}
+        {{ messages[0]['content'] | trim }}
+        {%- set loop_messages = messages[1:] -%}
+    {%- endif -%}
+    {{ '<turn|>
+' }}
+{%- 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 -%}
+    {{ '<|turn>' + role + '
+' }}
+    {%- if message['content'] is string -%}
+        {%- if role == "model" -%}
+            {{ strip_thinking(message['content']) }}
+        {%- else -%}
+            {{ message['content'] | trim }}
+        {%- endif -%}
+    {%- elif message['content'] is iterable -%}
+        {%- for item in message['content'] -%}
+            {%- if item['type'] == 'audio' -%}
+                {{ '<|audio|>' }}
+            {%- elif item['type'] == 'image' -%}
+                {{ '<|image|>' }}
+            {%- elif item['type'] == 'video' -%}
+                {{ '<|video|>' }}
+            {%- elif item['type'] == 'text' -%}
+                {%- if role == "model" -%}
+                    {{ strip_thinking(item['text']) }}
+                {%- else -%}
+                    {{ item['text'] | trim }}
+                {%- endif -%}
+            {%- endif -%}
+        {%- endfor -%}
+    {%- else -%}
+        {{ raise_exception("Invalid content type") }}
+    {%- endif -%}
+    {{ '<turn|>
+' }}
+{%- endfor -%}
+{%- if add_generation_prompt -%}
+    {{'<|turn>model
+'}}
+{%- endif -%}

gemma4_e2b_out/checkpoint-1000/optimizer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a2575ed92e8989cf22e760c7aeb0346eb2d9dcbca70e4e57e5e7dd63ffb23173
+size 98628429

gemma4_e2b_out/checkpoint-1000/rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f1d565802a8e26c4e8a31328752b7a7fdc186d9401aa008e65697d0ad8c22e33
+size 14645

gemma4_e2b_out/checkpoint-1000/scaler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:14ae2a2128444abab378aa06c09a61a84665f758fcc19fc46f5789b0bc1b5665
+size 1383

gemma4_e2b_out/checkpoint-1000/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ada16671df03ce2c4c7b2196578593f96e5a9638c91cc68f984858ec0e816498
+size 1465

gemma4_e2b_out/checkpoint-1000/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cc8d3a0ce36466ccc1278bf987df5f71db1719b9ca6b4118264f45cb627bfe0f
+size 32169626

gemma4_e2b_out/checkpoint-1000/tokenizer_config.json

@@ -0,0 +1,289 @@
+{
+  "audio_token": "<|audio|>",
+  "backend": "tokenizers",
+  "boa_token": "<|audio>",
+  "boi_token": "<|image>",
+  "bos_token": "<bos>",
+  "eoa_token": "<audio|>",
+  "eoc_token": "<channel|>",
+  "eoi_token": "<image|>",
+  "eos_token": "<eos>",
+  "eot_token": "<turn|>",
+  "escape_token": "<|\"|>",
+  "etc_token": "<tool_call|>",
+  "etd_token": "<tool|>",
+  "etr_token": "<tool_response|>",
+  "extra_special_tokens": [
+    "<|video|>"
+  ],
+  "image_token": "<|image|>",
+  "is_local": false,
+  "mask_token": "<mask>",
+  "model_max_length": 131072,
+  "model_specific_special_tokens": {
+    "audio_token": "<|audio|>",
+    "boa_token": "<|audio>",
+    "boi_token": "<|image>",
+    "eoa_token": "<audio|>",
+    "eoc_token": "<channel|>",
+    "eoi_token": "<image|>",
+    "eot_token": "<turn|>",
+    "escape_token": "<|\"|>",
+    "etc_token": "<tool_call|>",
+    "etd_token": "<tool|>",
+    "etr_token": "<tool_response|>",
+    "image_token": "<|image|>",
+    "soc_token": "<|channel>",
+    "sot_token": "<|turn>",
+    "stc_token": "<|tool_call>",
+    "std_token": "<|tool>",
+    "str_token": "<|tool_response>",
+    "think_token": "<|think|>"
+  },
+  "pad_token": "<pad>",
+  "padding_side": "right",
+  "processor_class": "Gemma4Processor",
+  "response_schema": {
+    "properties": {
+      "content": {
+        "type": "string"
+      },
+      "role": {
+        "const": "assistant"
+      },
+      "thinking": {
+        "type": "string"
+      },
+      "tool_calls": {
+        "items": {
+          "properties": {
+            "function": {
+              "properties": {
+                "arguments": {
+                  "additionalProperties": {},
+                  "type": "object",
+                  "x-parser": "gemma4-tool-call"
+                },
+                "name": {
+                  "type": "string"
+                }
+              },
+              "type": "object",
+              "x-regex": "call\\:(?P<name>\\w+)(?P<arguments>\\{.*\\})"
+            },
+            "type": {
+              "const": "function"
+            }
+          },
+          "type": "object"
+        },
+        "type": "array",
+        "x-regex-iterator": "<\\|tool_call>(.*?)<tool_call\\|>"
+      }
+    },
+    "type": "object",
+    "x-regex": "(\\<\\|channel\\>thought\\n(?P<thinking>.*?)\\<channel\\|\\>)?(?P<tool_calls>\\<\\|tool_call\\>.*\\<tool_call\\|\\>)?(?P<content>(?:(?!\\<turn\\|\\>)(?!\\<\\|tool_response\\>).)+)?(?:\\<turn\\|\\>|\\<\\|tool_response\\>)?"
+  },
+  "soc_token": "<|channel>",
+  "sot_token": "<|turn>",
+  "stc_token": "<|tool_call>",
+  "std_token": "<|tool>",
+  "str_token": "<|tool_response>",
+  "think_token": "<|think|>",
+  "tokenizer_class": "GemmaTokenizer",
+  "unk_token": "<unk>",
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<pad>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "1": {
+      "content": "<eos>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "2": {
+      "content": "<bos>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "3": {
+      "content": "<unk>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "4": {
+      "content": "<mask>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "46": {
+      "content": "<|tool>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "47": {
+      "content": "<tool|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "48": {
+      "content": "<|tool_call>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "49": {
+      "content": "<tool_call|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "50": {
+      "content": "<|tool_response>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "51": {
+      "content": "<tool_response|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "52": {
+      "content": "<|\"|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "98": {
+      "content": "<|think|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "100": {
+      "content": "<|channel>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "101": {
+      "content": "<channel|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "105": {
+      "content": "<|turn>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "106": {
+      "content": "<turn|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "255999": {
+      "content": "<|image>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "256000": {
+      "content": "<|audio>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "258880": {
+      "content": "<|image|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "258881": {
+      "content": "<|audio|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "258882": {
+      "content": "<image|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "258883": {
+      "content": "<audio|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "258884": {
+      "content": "<|video|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    }
+  }
+}

gemma4_e2b_out/checkpoint-1000/trainer_state.json

@@ -0,0 +1,734 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.024417639302632223,
+  "eval_steps": 500,
+  "global_step": 1000,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.00024417639302632224,
+      "grad_norm": 0.48828125,
+      "learning_rate": 1.8e-05,
+      "loss": 1.1311161994934082,
+      "step": 10
+    },
+    {
+      "epoch": 0.0004883527860526445,
+      "grad_norm": 1.046875,
+      "learning_rate": 3.8e-05,
+      "loss": 1.2175333976745606,
+      "step": 20
+    },
+    {
+      "epoch": 0.0007325291790789666,
+      "grad_norm": 0.55859375,
+      "learning_rate": 5.8e-05,
+      "loss": 0.8916121482849121,
+      "step": 30
+    },
+    {
+      "epoch": 0.000976705572105289,
+      "grad_norm": 0.349609375,
+      "learning_rate": 7.800000000000001e-05,
+      "loss": 0.588713550567627,
+      "step": 40
+    },
+    {
+      "epoch": 0.0012208819651316112,
+      "grad_norm": 0.2431640625,
+      "learning_rate": 9.8e-05,
+      "loss": 0.5149059772491456,
+      "step": 50
+    },
+    {
+      "epoch": 0.0014650583581579332,
+      "grad_norm": 0.1484375,
+      "learning_rate": 0.000118,
+      "loss": 0.4876229286193848,
+      "step": 60
+    },
+    {
+      "epoch": 0.0017092347511842554,
+      "grad_norm": 0.10400390625,
+      "learning_rate": 0.000138,
+      "loss": 0.4492696762084961,
+      "step": 70
+    },
+    {
+      "epoch": 0.001953411144210578,
+      "grad_norm": 0.0966796875,
+      "learning_rate": 0.00015800000000000002,
+      "loss": 0.4119694232940674,
+      "step": 80
+    },
+    {
+      "epoch": 0.0021975875372369,
+      "grad_norm": 0.1025390625,
+      "learning_rate": 0.00017800000000000002,
+      "loss": 0.42351255416870115,
+      "step": 90
+    },
+    {
+      "epoch": 0.0024417639302632224,
+      "grad_norm": 0.73828125,
+      "learning_rate": 0.00019800000000000002,
+      "loss": 0.413313627243042,
+      "step": 100
+    },
+    {
+      "epoch": 0.0026859403232895444,
+      "grad_norm": 0.11376953125,
+      "learning_rate": 0.00019995065603657316,
+      "loss": 0.4356351852416992,
+      "step": 110
+    },
+    {
+      "epoch": 0.0029301167163158664,
+      "grad_norm": 0.1376953125,
+      "learning_rate": 0.000199780146829205,
+      "loss": 0.3644850492477417,
+      "step": 120
+    },
+    {
+      "epoch": 0.003174293109342189,
+      "grad_norm": 0.07421875,
+      "learning_rate": 0.00019948807088287883,
+      "loss": 0.392529559135437,
+      "step": 130
+    },
+    {
+      "epoch": 0.003418469502368511,
+      "grad_norm": 0.11279296875,
+      "learning_rate": 0.00019907478404714436,
+      "loss": 0.42130446434020996,
+      "step": 140
+    },
+    {
+      "epoch": 0.0036626458953948333,
+      "grad_norm": 0.12109375,
+      "learning_rate": 0.00019854078984834903,
+      "loss": 0.4250969409942627,
+      "step": 150
+    },
+    {
+      "epoch": 0.003906822288421156,
+      "grad_norm": 0.10693359375,
+      "learning_rate": 0.0001978867388761685,
+      "loss": 0.3913136005401611,
+      "step": 160
+    },
+    {
+      "epoch": 0.004150998681447478,
+      "grad_norm": 0.08984375,
+      "learning_rate": 0.00019711342799096361,
+      "loss": 0.40640673637390134,
+      "step": 170
+    },
+    {
+      "epoch": 0.0043951750744738,
+      "grad_norm": 0.1005859375,
+      "learning_rate": 0.00019622179935292855,
+      "loss": 0.3612894773483276,
+      "step": 180
+    },
+    {
+      "epoch": 0.004639351467500122,
+      "grad_norm": 0.10302734375,
+      "learning_rate": 0.00019521293927421388,
+      "loss": 0.3791630268096924,
+      "step": 190
+    },
+    {
+      "epoch": 0.004883527860526445,
+      "grad_norm": 0.05712890625,
+      "learning_rate": 0.00019408807689542257,
+      "loss": 0.3566859245300293,
+      "step": 200
+    },
+    {
+      "epoch": 0.005127704253552767,
+      "grad_norm": 0.0830078125,
+      "learning_rate": 0.00019284858268809137,
+      "loss": 0.3679236888885498,
+      "step": 210
+    },
+    {
+      "epoch": 0.005371880646579089,
+      "grad_norm": 0.10888671875,
+      "learning_rate": 0.0001914959667849825,
+      "loss": 0.38951241970062256,
+      "step": 220
+    },
+    {
+      "epoch": 0.005616057039605411,
+      "grad_norm": 0.087890625,
+      "learning_rate": 0.00019003187714021938,
+      "loss": 0.380203104019165,
+      "step": 230
+    },
+    {
+      "epoch": 0.005860233432631733,
+      "grad_norm": 0.11767578125,
+      "learning_rate": 0.0001884580975215084,
+      "loss": 0.3717670202255249,
+      "step": 240
+    },
+    {
+      "epoch": 0.006104409825658056,
+      "grad_norm": 0.07275390625,
+      "learning_rate": 0.00018677654533689287,
+      "loss": 0.3654949426651001,
+      "step": 250
+    },
+    {
+      "epoch": 0.006348586218684378,
+      "grad_norm": 0.11572265625,
+      "learning_rate": 0.00018498926929868642,
+      "loss": 0.32950897216796876,
+      "step": 260
+    },
+    {
+      "epoch": 0.0065927626117107,
+      "grad_norm": 0.08984375,
+      "learning_rate": 0.00018309844692743283,
+      "loss": 0.35263752937316895,
+      "step": 270
+    },
+    {
+      "epoch": 0.006836939004737022,
+      "grad_norm": 0.111328125,
+      "learning_rate": 0.00018110638189893267,
+      "loss": 0.37081763744354246,
+      "step": 280
+    },
+    {
+      "epoch": 0.007081115397763345,
+      "grad_norm": 0.0625,
+      "learning_rate": 0.00017901550123756906,
+      "loss": 0.34837267398834226,
+      "step": 290
+    },
+    {
+      "epoch": 0.007325291790789667,
+      "grad_norm": 0.078125,
+      "learning_rate": 0.00017682835235935236,
+      "loss": 0.37078888416290284,
+      "step": 300
+    },
+    {
+      "epoch": 0.007569468183815989,
+      "grad_norm": 0.08447265625,
+      "learning_rate": 0.00017454759996828623,
+      "loss": 0.3861753702163696,
+      "step": 310
+    },
+    {
+      "epoch": 0.007813644576842312,
+      "grad_norm": 0.10498046875,
+      "learning_rate": 0.00017217602280983623,
+      "loss": 0.32866339683532714,
+      "step": 320
+    },
+    {
+      "epoch": 0.008057820969868634,
+      "grad_norm": 0.07568359375,
+      "learning_rate": 0.00016971651028545648,
+      "loss": 0.4176007270812988,
+      "step": 330
+    },
+    {
+      "epoch": 0.008301997362894956,
+      "grad_norm": 0.07666015625,
+      "learning_rate": 0.00016717205893229903,
+      "loss": 0.32620184421539306,
+      "step": 340
+    },
+    {
+      "epoch": 0.008546173755921278,
+      "grad_norm": 0.0732421875,
+      "learning_rate": 0.00016454576877239507,
+      "loss": 0.36540043354034424,
+      "step": 350
+    },
+    {
+      "epoch": 0.0087903501489476,
+      "grad_norm": 0.057861328125,
+      "learning_rate": 0.0001618408395357554,
+      "loss": 0.3834134578704834,
+      "step": 360
+    },
+    {
+      "epoch": 0.009034526541973922,
+      "grad_norm": 0.0771484375,
+      "learning_rate": 0.00015906056676199255,
+      "loss": 0.3537560701370239,
+      "step": 370
+    },
+    {
+      "epoch": 0.009278702935000244,
+      "grad_norm": 0.1044921875,
+      "learning_rate": 0.00015620833778521307,
+      "loss": 0.39911091327667236,
+      "step": 380
+    },
+    {
+      "epoch": 0.009522879328026566,
+      "grad_norm": 0.0830078125,
+      "learning_rate": 0.000153287627607073,
+      "loss": 0.3524473667144775,
+      "step": 390
+    },
+    {
+      "epoch": 0.00976705572105289,
+      "grad_norm": 0.09619140625,
+      "learning_rate": 0.00015030199466302353,
+      "loss": 0.37404372692108157,
+      "step": 400
+    },
+    {
+      "epoch": 0.010011232114079211,
+      "grad_norm": 0.0859375,
+      "learning_rate": 0.00014725507648690543,
+      "loss": 0.3956392765045166,
+      "step": 410
+    },
+    {
+      "epoch": 0.010255408507105533,
+      "grad_norm": 0.06787109375,
+      "learning_rate": 0.00014415058527917452,
+      "loss": 0.375267767906189,
+      "step": 420
+    },
+    {
+      "epoch": 0.010499584900131856,
+      "grad_norm": 0.1181640625,
+      "learning_rate": 0.00014099230338415728,
+      "loss": 0.3407683610916138,
+      "step": 430
+    },
+    {
+      "epoch": 0.010743761293158178,
+      "grad_norm": 0.068359375,
+      "learning_rate": 0.00013778407868184672,
+      "loss": 0.3405567407608032,
+      "step": 440
+    },
+    {
+      "epoch": 0.0109879376861845,
+      "grad_norm": 0.0810546875,
+      "learning_rate": 0.00013452981989985348,
+      "loss": 0.3458467960357666,
+      "step": 450
+    },
+    {
+      "epoch": 0.011232114079210822,
+      "grad_norm": 0.0732421875,
+      "learning_rate": 0.00013123349185122327,
+      "loss": 0.37671115398406985,
+      "step": 460
+    },
+    {
+      "epoch": 0.011476290472237144,
+      "grad_norm": 0.07275390625,
+      "learning_rate": 0.00012789911060392294,
+      "loss": 0.3458314180374146,
+      "step": 470
+    },
+    {
+      "epoch": 0.011720466865263466,
+      "grad_norm": 0.091796875,
+      "learning_rate": 0.00012453073858788026,
+      "loss": 0.3307004690170288,
+      "step": 480
+    },
+    {
+      "epoch": 0.01196464325828979,
+      "grad_norm": 0.0791015625,
+      "learning_rate": 0.00012113247964553888,
+      "loss": 0.333436393737793,
+      "step": 490
+    },
+    {
+      "epoch": 0.012208819651316111,
+      "grad_norm": 0.09814453125,
+      "learning_rate": 0.00011770847403195834,
+      "loss": 0.3636301517486572,
+      "step": 500
+    },
+    {
+      "epoch": 0.012452996044342433,
+      "grad_norm": 0.181640625,
+      "learning_rate": 0.00011426289337055119,
+      "loss": 0.3728192329406738,
+      "step": 510
+    },
+    {
+      "epoch": 0.012697172437368755,
+      "grad_norm": 0.134765625,
+      "learning_rate": 0.0001107999355706023,
+      "loss": 0.3645843505859375,
+      "step": 520
+    },
+    {
+      "epoch": 0.012941348830395077,
+      "grad_norm": 0.1201171875,
+      "learning_rate": 0.00010732381971276318,
+      "loss": 0.3898338556289673,
+      "step": 530
+    },
+    {
+      "epoch": 0.0131855252234214,
+      "grad_norm": 0.09375,
+      "learning_rate": 0.00010383878090875201,
+      "loss": 0.36258883476257325,
+      "step": 540
+    },
+    {
+      "epoch": 0.013429701616447721,
+      "grad_norm": 0.08740234375,
+      "learning_rate": 0.00010034906514152238,
+      "loss": 0.35727477073669434,
+      "step": 550
+    },
+    {
+      "epoch": 0.013673878009474043,
+      "grad_norm": 0.0732421875,
+      "learning_rate": 9.685892409218717e-05,
+      "loss": 0.3593540906906128,
+      "step": 560
+    },
+    {
+      "epoch": 0.013918054402500366,
+      "grad_norm": 0.10595703125,
+      "learning_rate": 9.337260996000002e-05,
+      "loss": 0.3762841701507568,
+      "step": 570
+    },
+    {
+      "epoch": 0.01416223079552669,
+      "grad_norm": 0.08984375,
+      "learning_rate": 8.989437028170537e-05,
+      "loss": 0.336023736000061,
+      "step": 580
+    },
+    {
+      "epoch": 0.014406407188553011,
+      "grad_norm": 0.09814453125,
+      "learning_rate": 8.642844275656957e-05,
+      "loss": 0.33524041175842284,
+      "step": 590
+    },
+    {
+      "epoch": 0.014650583581579333,
+      "grad_norm": 0.09228515625,
+      "learning_rate": 8.297905008339677e-05,
+      "loss": 0.392057204246521,
+      "step": 600
+    },
+    {
+      "epoch": 0.014894759974605655,
+      "grad_norm": 0.111328125,
+      "learning_rate": 7.955039481582097e-05,
+      "loss": 0.3882176399230957,
+      "step": 610
+    },
+    {
+      "epoch": 0.015138936367631977,
+      "grad_norm": 0.06640625,
+      "learning_rate": 7.614665424214193e-05,
+      "loss": 0.38143367767333985,
+      "step": 620
+    },
+    {
+      "epoch": 0.0153831127606583,
+      "grad_norm": 0.0859375,
+      "learning_rate": 7.277197529594257e-05,
+      "loss": 0.352864146232605,
+      "step": 630
+    },
+    {
+      "epoch": 0.015627289153684623,
+      "grad_norm": 0.07666015625,
+      "learning_rate": 6.943046950368944e-05,
+      "loss": 0.35377163887023927,
+      "step": 640
+    },
+    {
+      "epoch": 0.015871465546710945,
+      "grad_norm": 0.076171875,
+      "learning_rate": 6.612620797547087e-05,
+      "loss": 0.36560447216033937,
+      "step": 650
+    },
+    {
+      "epoch": 0.016115641939737267,
+      "grad_norm": 0.1142578125,
+      "learning_rate": 6.286321644497655e-05,
+      "loss": 0.3602383375167847,
+      "step": 660
+    },
+    {
+      "epoch": 0.01635981833276359,
+      "grad_norm": 0.0654296875,
+      "learning_rate": 5.964547036476099e-05,
+      "loss": 0.3708587646484375,
+      "step": 670
+    },
+    {
+      "epoch": 0.01660399472578991,
+      "grad_norm": 0.07080078125,
+      "learning_rate": 5.647689006276726e-05,
+      "loss": 0.33884291648864745,
+      "step": 680
+    },
+    {
+      "epoch": 0.016848171118816233,
+      "grad_norm": 0.072265625,
+      "learning_rate": 5.33613359660109e-05,
+      "loss": 0.3408441781997681,
+      "step": 690
+    },
+    {
+      "epoch": 0.017092347511842555,
+      "grad_norm": 0.11572265625,
+      "learning_rate": 5.0302603897244474e-05,
+      "loss": 0.3734944581985474,
+      "step": 700
+    },
+    {
+      "epoch": 0.017336523904868877,
+      "grad_norm": 0.07080078125,
+      "learning_rate": 4.7304420450332244e-05,
+      "loss": 0.33618974685668945,
+      "step": 710
+    },
+    {
+      "epoch": 0.0175807002978952,
+      "grad_norm": 0.08349609375,
+      "learning_rate": 4.437043844996952e-05,
+      "loss": 0.3502551794052124,
+      "step": 720
+    },
+    {
+      "epoch": 0.01782487669092152,
+      "grad_norm": 0.076171875,
+      "learning_rate": 4.150423250127845e-05,
+      "loss": 0.3836493492126465,
+      "step": 730
+    },
+    {
+      "epoch": 0.018069053083947843,
+      "grad_norm": 0.1064453125,
+      "learning_rate": 3.8709294634702376e-05,
+      "loss": 0.33480191230773926,
+      "step": 740
+    },
+    {
+      "epoch": 0.018313229476974165,
+      "grad_norm": 0.0693359375,
+      "learning_rate": 3.5989030051504434e-05,
+      "loss": 0.3809062480926514,
+      "step": 750
+    },
+    {
+      "epoch": 0.018557405870000487,
+      "grad_norm": 0.1064453125,
+      "learning_rate": 3.334675297505476e-05,
+      "loss": 0.3911257266998291,
+      "step": 760
+    },
+    {
+      "epoch": 0.01880158226302681,
+      "grad_norm": 0.11083984375,
+      "learning_rate": 3.078568261295933e-05,
+      "loss": 0.38447113037109376,
+      "step": 770
+    },
+    {
+      "epoch": 0.01904575865605313,
+      "grad_norm": 0.10693359375,
+      "learning_rate": 2.8308939234951726e-05,
+      "loss": 0.3653350830078125,
+      "step": 780
+    },
+    {
+      "epoch": 0.019289935049079453,
+      "grad_norm": 0.078125,
+      "learning_rate": 2.5919540371325e-05,
+      "loss": 0.3945863485336304,
+      "step": 790
+    },
+    {
+      "epoch": 0.01953411144210578,
+      "grad_norm": 0.11376953125,
+      "learning_rate": 2.362039713653581e-05,
+      "loss": 0.39480888843536377,
+      "step": 800
+    },
+    {
+      "epoch": 0.0197782878351321,
+      "grad_norm": 0.0771484375,
+      "learning_rate": 2.1414310682459802e-05,
+      "loss": 0.289493727684021,
+      "step": 810
+    },
+    {
+      "epoch": 0.020022464228158423,
+      "grad_norm": 0.0849609375,
+      "learning_rate": 1.930396878561983e-05,
+      "loss": 0.3639736890792847,
+      "step": 820
+    },
+    {
+      "epoch": 0.020266640621184745,
+      "grad_norm": 0.1005859375,
+      "learning_rate": 1.7291942572543807e-05,
+      "loss": 0.36269190311431887,
+      "step": 830
+    },
+    {
+      "epoch": 0.020510817014211067,
+      "grad_norm": 0.10009765625,
+      "learning_rate": 1.538068338724361e-05,
+      "loss": 0.3807779312133789,
+      "step": 840
+    },
+    {
+      "epoch": 0.02075499340723739,
+      "grad_norm": 0.10205078125,
+      "learning_rate": 1.3572519804629536e-05,
+      "loss": 0.39589340686798097,
+      "step": 850
+    },
+    {
+      "epoch": 0.02099916980026371,
+      "grad_norm": 0.09521484375,
+      "learning_rate": 1.1869654793500784e-05,
+      "loss": 0.393012261390686,
+      "step": 860
+    },
+    {
+      "epoch": 0.021243346193290033,
+      "grad_norm": 0.07470703125,
+      "learning_rate": 1.0274163032567163e-05,
+      "loss": 0.3827983379364014,
+      "step": 870
+    },
+    {
+      "epoch": 0.021487522586316355,
+      "grad_norm": 0.08251953125,
+      "learning_rate": 8.787988382772705e-06,
+      "loss": 0.3293968915939331,
+      "step": 880
+    },
+    {
+      "epoch": 0.021731698979342677,
+      "grad_norm": 0.0693359375,
+      "learning_rate": 7.412941519000527e-06,
+      "loss": 0.3682440519332886,
+      "step": 890
+    },
+    {
+      "epoch": 0.021975875372369,
+      "grad_norm": 0.09228515625,
+      "learning_rate": 6.1506977240444074e-06,
+      "loss": 0.3646101474761963,
+      "step": 900
+    },
+    {
+      "epoch": 0.02222005176539532,
+      "grad_norm": 0.11669921875,
+      "learning_rate": 5.002794847534764e-06,
+      "loss": 0.38763275146484377,
+      "step": 910
+    },
+    {
+      "epoch": 0.022464228158421643,
+      "grad_norm": 0.1142578125,
+      "learning_rate": 3.970631432305694e-06,
+      "loss": 0.34759066104888914,
+      "step": 920
+    },
+    {
+      "epoch": 0.022708404551447965,
+      "grad_norm": 0.07177734375,
+      "learning_rate": 3.0554650104861136e-06,
+      "loss": 0.35361154079437257,
+      "step": 930
+    },
+    {
+      "epoch": 0.022952580944474287,
+      "grad_norm": 0.1259765625,
+      "learning_rate": 2.2584105713904125e-06,
+      "loss": 0.36288425922393797,
+      "step": 940
+    },
+    {
+      "epoch": 0.02319675733750061,
+      "grad_norm": 0.099609375,
+      "learning_rate": 1.580439203075812e-06,
+      "loss": 0.3696069002151489,
+      "step": 950
+    },
+    {
+      "epoch": 0.02344093373052693,
+      "grad_norm": 0.06982421875,
+      "learning_rate": 1.0223769092211012e-06,
+      "loss": 0.3502499103546143,
+      "step": 960
+    },
+    {
+      "epoch": 0.023685110123553253,
+      "grad_norm": 0.072265625,
+      "learning_rate": 5.849036027684606e-07,
+      "loss": 0.3617737293243408,
+      "step": 970
+    },
+    {
+      "epoch": 0.02392928651657958,
+      "grad_norm": 0.12109375,
+      "learning_rate": 2.685522775541904e-07,
+      "loss": 0.38356838226318357,
+      "step": 980
+    },
+    {
+      "epoch": 0.0241734629096059,
+      "grad_norm": 0.09130859375,
+      "learning_rate": 7.370835893788508e-08,
+      "loss": 0.38213505744934084,
+      "step": 990
+    },
+    {
+      "epoch": 0.024417639302632223,
+      "grad_norm": 0.07568359375,
+      "learning_rate": 6.092342209607083e-10,
+      "loss": 0.3387150287628174,
+      "step": 1000
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 1000,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "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": 9.089382042631987e+16,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}

gemma4_e2b_out/checkpoint-1000/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:01c81c440c08fc4b233b9ce62ac6c2e3f9a0369823edad1a08fde166c4472b8d
+size 5841

gemma4_gguf/chat_template.jinja

@@ -0,0 +1,70 @@
+{{ bos_token }}{%- macro strip_thinking(text) -%}
+    {%- set ns = namespace(result='') -%}
+    {%- for part in text.split('<channel|>') -%}
+        {%- if '<|channel>' in part -%}
+            {%- set ns.result = ns.result + part.split('<|channel>')[0] -%}
+        {%- else -%}
+            {%- set ns.result = ns.result + part -%}
+        {%- endif -%}
+    {%- endfor -%}
+    {{- ns.result | trim -}}
+{%- endmacro -%}
+{%- set thinking = enable_thinking is defined and enable_thinking -%}
+{%- set loop_messages = messages -%}
+{%- if messages[0]['role'] in ['system', 'developer'] or thinking -%}
+    {{ '<|turn>system
+' }}
+    {%- if thinking -%}
+        {{ '<|think|>
+' }}
+    {%- endif -%}
+    {%- if messages[0]['role'] in ['system', 'developer'] -%}
+        {{ messages[0]['content'] | trim }}
+        {%- set loop_messages = messages[1:] -%}
+    {%- endif -%}
+    {{ '<turn|>
+' }}
+{%- 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 -%}
+    {{ '<|turn>' + role + '
+' }}
+    {%- if message['content'] is string -%}
+        {%- if role == "model" -%}
+            {{ strip_thinking(message['content']) }}
+        {%- else -%}
+            {{ message['content'] | trim }}
+        {%- endif -%}
+    {%- elif message['content'] is iterable -%}
+        {%- for item in message['content'] -%}
+            {%- if item['type'] == 'audio' -%}
+                {{ '<|audio|>' }}
+            {%- elif item['type'] == 'image' -%}
+                {{ '<|image|>' }}
+            {%- elif item['type'] == 'video' -%}
+                {{ '<|video|>' }}
+            {%- elif item['type'] == 'text' -%}
+                {%- if role == "model" -%}
+                    {{ strip_thinking(item['text']) }}
+                {%- else -%}
+                    {{ item['text'] | trim }}
+                {%- endif -%}
+            {%- endif -%}
+        {%- endfor -%}
+    {%- else -%}
+        {{ raise_exception("Invalid content type") }}
+    {%- endif -%}
+    {{ '<turn|>
+' }}
+{%- endfor -%}
+{%- if add_generation_prompt -%}
+    {{'<|turn>model
+'}}
+{%- endif -%}

gemma4_gguf/config.json

@@ -0,0 +1,193 @@
+{
+    "architectures": [
+        "Gemma4ForConditionalGeneration"
+    ],
+    "audio_config": {
+        "_name_or_path": "",
+        "architectures": null,
+        "attention_chunk_size": 12,
+        "attention_context_left": 13,
+        "attention_context_right": 0,
+        "attention_invalid_logits_value": -1000000000.0,
+        "attention_logit_cap": 50.0,
+        "chunk_size_feed_forward": 0,
+        "conv_kernel_size": 5,
+        "torch_dtype": "float16",
+        "gradient_clipping": 10000000000.0,
+        "hidden_act": "silu",
+        "hidden_size": 1024,
+        "id2label": {
+            "0": "LABEL_0",
+            "1": "LABEL_1"
+        },
+        "initializer_range": 0.02,
+        "is_encoder_decoder": false,
+        "label2id": {
+            "LABEL_0": 0,
+            "LABEL_1": 1
+        },
+        "model_type": "gemma4_audio",
+        "num_attention_heads": 8,
+        "num_hidden_layers": 12,
+        "output_attentions": false,
+        "output_hidden_states": false,
+        "output_proj_dims": 1536,
+        "problem_type": null,
+        "residual_weight": 0.5,
+        "return_dict": true,
+        "rms_norm_eps": 1e-06,
+        "subsampling_conv_channels": [
+            128,
+            32
+        ],
+        "use_clipped_linears": true
+    },
+    "audio_token_id": 258881,
+    "boa_token_id": 256000,
+    "boi_token_id": 255999,
+    "bos_token_id": 2,
+    "torch_dtype": "float16",
+    "eoa_token_id": 258883,
+    "eoa_token_index": 258883,
+    "eoi_token_id": 258882,
+    "eos_token_id": 1,
+    "image_token_id": 258880,
+    "initializer_range": 0.02,
+    "model_name": "unsloth/gemma-4-e2b-it-unsloth-bnb-4bit",
+    "model_type": "gemma4",
+    "pad_token_id": 0,
+    "text_config": {
+        "attention_bias": false,
+        "attention_dropout": 0.0,
+        "attention_k_eq_v": false,
+        "bos_token_id": 2,
+        "torch_dtype": "float16",
+        "enable_moe_block": false,
+        "eos_token_id": 1,
+        "expert_intermediate_size": null,
+        "final_logit_softcapping": 30.0,
+        "global_head_dim": 512,
+        "head_dim": 256,
+        "hidden_activation": "gelu_pytorch_tanh",
+        "hidden_size": 1536,
+        "hidden_size_per_layer_input": 256,
+        "initializer_range": 0.02,
+        "intermediate_size": 6144,
+        "layer_types": [
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "full_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "full_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "full_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "full_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "full_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "full_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "sliding_attention",
+            "full_attention"
+        ],
+        "max_position_embeddings": 131072,
+        "model_type": "gemma4_text",
+        "moe_intermediate_size": null,
+        "num_attention_heads": 8,
+        "num_experts": null,
+        "num_global_key_value_heads": null,
+        "num_hidden_layers": 35,
+        "num_key_value_heads": 1,
+        "num_kv_shared_layers": 20,
+        "pad_token_id": 0,
+        "rms_norm_eps": 1e-06,
+        "rope_parameters": {
+            "full_attention": {
+                "partial_rotary_factor": 0.25,
+                "rope_theta": 1000000.0,
+                "rope_type": "proportional"
+            },
+            "sliding_attention": {
+                "rope_theta": 10000.0,
+                "rope_type": "default"
+            }
+        },
+        "sliding_window": 512,
+        "tie_word_embeddings": true,
+        "top_k_experts": null,
+        "use_bidirectional_attention": null,
+        "use_cache": true,
+        "use_double_wide_mlp": true,
+        "vocab_size": 262144,
+        "vocab_size_per_layer_input": 262144
+    },
+    "tie_word_embeddings": true,
+    "unsloth_fixed": true,
+    "unsloth_version": "2026.6.9",
+    "use_cache": false,
+    "video_token_id": 258884,
+    "vision_config": {
+        "_name_or_path": "",
+        "architectures": null,
+        "attention_bias": false,
+        "attention_dropout": 0.0,
+        "chunk_size_feed_forward": 0,
+        "default_output_length": 280,
+        "torch_dtype": "float16",
+        "global_head_dim": 64,
+        "head_dim": 64,
+        "hidden_activation": "gelu_pytorch_tanh",
+        "hidden_size": 768,
+        "id2label": {
+            "0": "LABEL_0",
+            "1": "LABEL_1"
+        },
+        "initializer_range": 0.02,
+        "intermediate_size": 3072,
+        "is_encoder_decoder": false,
+        "label2id": {
+            "LABEL_0": 0,
+            "LABEL_1": 1
+        },
+        "max_position_embeddings": 131072,
+        "model_type": "gemma4_vision",
+        "num_attention_heads": 12,
+        "num_hidden_layers": 16,
+        "num_key_value_heads": 12,
+        "output_attentions": false,
+        "output_hidden_states": false,
+        "patch_size": 16,
+        "pooling_kernel_size": 3,
+        "position_embedding_size": 10240,
+        "problem_type": null,
+        "return_dict": true,
+        "rms_norm_eps": 1e-06,
+        "rope_parameters": {
+            "rope_theta": 100.0,
+            "rope_type": "default"
+        },
+        "standardize": false,
+        "use_clipped_linears": true
+    },
+    "vision_soft_tokens_per_image": 280
+}

gemma4_gguf/generation_config.json

@@ -0,0 +1,14 @@
+{
+  "bos_token_id": 2,
+  "do_sample": true,
+  "eos_token_id": [
+    1,
+    106,
+    50
+  ],
+  "pad_token_id": 0,
+  "temperature": 1.0,
+  "top_k": 64,
+  "top_p": 0.95,
+  "transformers_version": "5.5.0"
+}

gemma4_gguf/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:440203cafd5bac689bae28d22382b2bc67d5a39849b590cd88e862c191185d5c
+size 10246621918

gemma4_gguf/temp_split_0de139d7_000.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9ebcea1985a201703cd702b4bb5e3cfc4efddaa4e2b4c58fce3703baa456a11e
+size 1422130776

gemma4_gguf/temp_split_0de139d7_001.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2ea02bad44856b48a53786446e9d32217ea9c377b26d16032919b8785c2b87be
+size 4697620648

gemma4_gguf/temp_split_0de139d7_002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:85cebd4f098a5d61a725613814958869c645b022e6a7c083e69404732cd541b4
+size 1585731424

gemma4_gguf/temp_split_0de139d7_003.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e169e6e4b22e2ab8787a697a0cb82391d5cd15dccdc183ee0053976a4bbf2375
+size 1446899712

gemma4_gguf/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cc8d3a0ce36466ccc1278bf987df5f71db1719b9ca6b4118264f45cb627bfe0f
+size 32169626

gemma4_gguf/tokenizer_config.json

@@ -0,0 +1,290 @@
+{
+  "audio_token": "<|audio|>",
+  "backend": "tokenizers",
+  "boa_token": "<|audio>",
+  "boi_token": "<|image>",
+  "bos_token": "<bos>",
+  "eoa_token": "<audio|>",
+  "eoc_token": "<channel|>",
+  "eoi_token": "<image|>",
+  "eos_token": "<eos>",
+  "eot_token": "<turn|>",
+  "escape_token": "<|\"|>",
+  "etc_token": "<tool_call|>",
+  "etd_token": "<tool|>",
+  "etr_token": "<tool_response|>",
+  "extra_special_tokens": [
+    "<|video|>"
+  ],
+  "image_token": "<|image|>",
+  "is_local": false,
+  "mask_token": "<mask>",
+  "model_max_length": 131072,
+  "model_specific_special_tokens": {
+    "audio_token": "<|audio|>",
+    "boa_token": "<|audio>",
+    "boi_token": "<|image>",
+    "eoa_token": "<audio|>",
+    "eoc_token": "<channel|>",
+    "eoi_token": "<image|>",
+    "eot_token": "<turn|>",
+    "escape_token": "<|\"|>",
+    "etc_token": "<tool_call|>",
+    "etd_token": "<tool|>",
+    "etr_token": "<tool_response|>",
+    "image_token": "<|image|>",
+    "soc_token": "<|channel>",
+    "sot_token": "<|turn>",
+    "stc_token": "<|tool_call>",
+    "std_token": "<|tool>",
+    "str_token": "<|tool_response>",
+    "think_token": "<|think|>"
+  },
+  "pad_token": "<pad>",
+  "padding_side": "right",
+  "processor_class": "Gemma4Processor",
+  "response_schema": {
+    "properties": {
+      "content": {
+        "type": "string"
+      },
+      "role": {
+        "const": "assistant"
+      },
+      "thinking": {
+        "type": "string"
+      },
+      "tool_calls": {
+        "items": {
+          "properties": {
+            "function": {
+              "properties": {
+                "arguments": {
+                  "additionalProperties": {},
+                  "type": "object",
+                  "x-parser": "gemma4-tool-call"
+                },
+                "name": {
+                  "type": "string"
+                }
+              },
+              "type": "object",
+              "x-regex": "call\\:(?P<name>\\w+)(?P<arguments>\\{.*\\})"
+            },
+            "type": {
+              "const": "function"
+            }
+          },
+          "type": "object"
+        },
+        "type": "array",
+        "x-regex-iterator": "<\\|tool_call>(.*?)<tool_call\\|>"
+      }
+    },
+    "type": "object",
+    "x-regex": "(\\<\\|channel\\>thought\\n(?P<thinking>.*?)\\<channel\\|\\>)?(?P<tool_calls>\\<\\|tool_call\\>.*\\<tool_call\\|\\>)?(?P<content>(?:(?!\\<turn\\|\\>)(?!\\<\\|tool_response\\>).)+)?(?:\\<turn\\|\\>|\\<\\|tool_response\\>)?"
+  },
+  "soc_token": "<|channel>",
+  "sot_token": "<|turn>",
+  "stc_token": "<|tool_call>",
+  "std_token": "<|tool>",
+  "str_token": "<|tool_response>",
+  "think_token": "<|think|>",
+  "tokenizer_class": "GemmaTokenizer",
+  "unk_token": "<unk>",
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<pad>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "1": {
+      "content": "<eos>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "2": {
+      "content": "<bos>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "3": {
+      "content": "<unk>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "4": {
+      "content": "<mask>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "46": {
+      "content": "<|tool>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "47": {
+      "content": "<tool|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "48": {
+      "content": "<|tool_call>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "49": {
+      "content": "<tool_call|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "50": {
+      "content": "<|tool_response>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "51": {
+      "content": "<tool_response|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "52": {
+      "content": "<|\"|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "98": {
+      "content": "<|think|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "100": {
+      "content": "<|channel>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "101": {
+      "content": "<channel|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "105": {
+      "content": "<|turn>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "106": {
+      "content": "<turn|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "255999": {
+      "content": "<|image>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "256000": {
+      "content": "<|audio>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "258880": {
+      "content": "<|image|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "258881": {
+      "content": "<|audio|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "258882": {
+      "content": "<image|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "258883": {
+      "content": "<audio|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "258884": {
+      "content": "<|video|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    }
+  },
+  "chat_template": "{{ bos_token }}{%- macro strip_thinking(text) -%}\n    {%- set ns = namespace(result='') -%}\n    {%- for part in text.split('<channel|>') -%}\n        {%- if '<|channel>' in part -%}\n            {%- set ns.result = ns.result + part.split('<|channel>')[0] -%}\n        {%- else -%}\n            {%- set ns.result = ns.result + part -%}\n        {%- endif -%}\n    {%- endfor -%}\n    {{- ns.result | trim -}}\n{%- endmacro -%}\n{%- set thinking = enable_thinking is defined and enable_thinking -%}\n{%- set loop_messages = messages -%}\n{%- if messages[0]['role'] in ['system', 'developer'] or thinking -%}\n    {{ '<|turn>system\n' }}\n    {%- if thinking -%}\n        {{ '<|think|>\n' }}\n    {%- endif -%}\n    {%- if messages[0]['role'] in ['system', 'developer'] -%}\n        {{ messages[0]['content'] | trim }}\n        {%- set loop_messages = messages[1:] -%}\n    {%- endif -%}\n    {{ '<turn|>\n' }}\n{%- endif -%}\n{%- for message in loop_messages -%}\n    {%- if (message['role'] == 'user') != (loop.index0 % 2 == 0) -%}\n        {{ raise_exception(\"Conversation roles must alternate user/assistant/user/assistant/...\") }}\n    {%- endif -%}\n    {%- if (message['role'] == 'assistant') -%}\n        {%- set role = \"model\" -%}\n    {%- else -%}\n        {%- set role = message['role'] -%}\n    {%- endif -%}\n    {{ '<|turn>' + role + '\n' }}\n    {%- if message['content'] is string -%}\n        {%- if role == \"model\" -%}\n            {{ strip_thinking(message['content']) }}\n        {%- else -%}\n            {{ message['content'] | trim }}\n        {%- endif -%}\n    {%- elif message['content'] is iterable -%}\n        {%- for item in message['content'] -%}\n            {%- if item['type'] == 'audio' -%}\n                {{ '<|audio|>' }}\n            {%- elif item['type'] == 'image' -%}\n                {{ '<|image|>' }}\n            {%- elif item['type'] == 'video' -%}\n                {{ '<|video|>' }}\n            {%- elif item['type'] == 'text' -%}\n                {%- if role == \"model\" -%}\n                    {{ strip_thinking(item['text']) }}\n                {%- else -%}\n                    {{ item['text'] | trim }}\n                {%- endif -%}\n            {%- endif -%}\n        {%- endfor -%}\n    {%- else -%}\n        {{ raise_exception(\"Invalid content type\") }}\n    {%- endif -%}\n    {{ '<turn|>\n' }}\n{%- endfor -%}\n{%- if add_generation_prompt -%}\n    {{'<|turn>model\n'}}\n{%- endif -%}\n"
+}

lora_weights_final/README.md

@@ -0,0 +1,210 @@
+---
+base_model: unsloth/gemma-4-e2b-it-unsloth-bnb-4bit
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:unsloth/gemma-4-e2b-it-unsloth-bnb-4bit
+- lora
+- sft
+- transformers
+- trl
+- unsloth
+---
+
+# 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

lora_weights_final/adapter_config.json

@@ -0,0 +1,44 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "Gemma4ForConditionalGeneration",
+    "parent_library": "transformers.models.gemma4.modeling_gemma4",
+    "unsloth_fixed": true
+  },
+  "base_model_name_or_path": "unsloth/gemma-4-e2b-it-unsloth-bnb-4bit",
+  "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,
+  "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": 32,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": "(?:.*?(?:language|text).*?(?:self_attn|attention|attn|mixer|mlp|feed_forward|ffn|dense|mixer).*?(?:q_proj|k_proj|v_proj|o_proj|gate_proj|up_proj|down_proj))|(?:\\bmodel\\.layers\\.[\\d]{1,}\\.(?:self_attn|attention|attn|mixer|mlp|feed_forward|ffn|dense|mixer)\\.(?:(?:q_proj|k_proj|v_proj|o_proj|gate_proj|up_proj|down_proj)))",
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_bdlora": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

lora_weights_final/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f14e7d5c4c6cfe03e5f99ffea455059061c7edc6b443e909b0c236b0bc8913cb
+size 101424416

lora_weights_final/chat_template.jinja

@@ -0,0 +1,70 @@
+{{ bos_token }}{%- macro strip_thinking(text) -%}
+    {%- set ns = namespace(result='') -%}
+    {%- for part in text.split('<channel|>') -%}
+        {%- if '<|channel>' in part -%}
+            {%- set ns.result = ns.result + part.split('<|channel>')[0] -%}
+        {%- else -%}
+            {%- set ns.result = ns.result + part -%}
+        {%- endif -%}
+    {%- endfor -%}
+    {{- ns.result | trim -}}
+{%- endmacro -%}
+{%- set thinking = enable_thinking is defined and enable_thinking -%}
+{%- set loop_messages = messages -%}
+{%- if messages[0]['role'] in ['system', 'developer'] or thinking -%}
+    {{ '<|turn>system
+' }}
+    {%- if thinking -%}
+        {{ '<|think|>
+' }}
+    {%- endif -%}
+    {%- if messages[0]['role'] in ['system', 'developer'] -%}
+        {{ messages[0]['content'] | trim }}
+        {%- set loop_messages = messages[1:] -%}
+    {%- endif -%}
+    {{ '<turn|>
+' }}
+{%- 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 -%}
+    {{ '<|turn>' + role + '
+' }}
+    {%- if message['content'] is string -%}
+        {%- if role == "model" -%}
+            {{ strip_thinking(message['content']) }}
+        {%- else -%}
+            {{ message['content'] | trim }}
+        {%- endif -%}
+    {%- elif message['content'] is iterable -%}
+        {%- for item in message['content'] -%}
+            {%- if item['type'] == 'audio' -%}
+                {{ '<|audio|>' }}
+            {%- elif item['type'] == 'image' -%}
+                {{ '<|image|>' }}
+            {%- elif item['type'] == 'video' -%}
+                {{ '<|video|>' }}
+            {%- elif item['type'] == 'text' -%}
+                {%- if role == "model" -%}
+                    {{ strip_thinking(item['text']) }}
+                {%- else -%}
+                    {{ item['text'] | trim }}
+                {%- endif -%}
+            {%- endif -%}
+        {%- endfor -%}
+    {%- else -%}
+        {{ raise_exception("Invalid content type") }}
+    {%- endif -%}
+    {{ '<turn|>
+' }}
+{%- endfor -%}
+{%- if add_generation_prompt -%}
+    {{'<|turn>model
+'}}
+{%- endif -%}

lora_weights_final/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cc8d3a0ce36466ccc1278bf987df5f71db1719b9ca6b4118264f45cb627bfe0f
+size 32169626

lora_weights_final/tokenizer_config.json

@@ -0,0 +1,289 @@
+{
+  "audio_token": "<|audio|>",
+  "backend": "tokenizers",
+  "boa_token": "<|audio>",
+  "boi_token": "<|image>",
+  "bos_token": "<bos>",
+  "eoa_token": "<audio|>",
+  "eoc_token": "<channel|>",
+  "eoi_token": "<image|>",
+  "eos_token": "<eos>",
+  "eot_token": "<turn|>",
+  "escape_token": "<|\"|>",
+  "etc_token": "<tool_call|>",
+  "etd_token": "<tool|>",
+  "etr_token": "<tool_response|>",
+  "extra_special_tokens": [
+    "<|video|>"
+  ],
+  "image_token": "<|image|>",
+  "is_local": false,
+  "mask_token": "<mask>",
+  "model_max_length": 131072,
+  "model_specific_special_tokens": {
+    "audio_token": "<|audio|>",
+    "boa_token": "<|audio>",
+    "boi_token": "<|image>",
+    "eoa_token": "<audio|>",
+    "eoc_token": "<channel|>",
+    "eoi_token": "<image|>",
+    "eot_token": "<turn|>",
+    "escape_token": "<|\"|>",
+    "etc_token": "<tool_call|>",
+    "etd_token": "<tool|>",
+    "etr_token": "<tool_response|>",
+    "image_token": "<|image|>",
+    "soc_token": "<|channel>",
+    "sot_token": "<|turn>",
+    "stc_token": "<|tool_call>",
+    "std_token": "<|tool>",
+    "str_token": "<|tool_response>",
+    "think_token": "<|think|>"
+  },
+  "pad_token": "<pad>",
+  "padding_side": "right",
+  "processor_class": "Gemma4Processor",
+  "response_schema": {
+    "properties": {
+      "content": {
+        "type": "string"
+      },
+      "role": {
+        "const": "assistant"
+      },
+      "thinking": {
+        "type": "string"
+      },
+      "tool_calls": {
+        "items": {
+          "properties": {
+            "function": {
+              "properties": {
+                "arguments": {
+                  "additionalProperties": {},
+                  "type": "object",
+                  "x-parser": "gemma4-tool-call"
+                },
+                "name": {
+                  "type": "string"
+                }
+              },
+              "type": "object",
+              "x-regex": "call\\:(?P<name>\\w+)(?P<arguments>\\{.*\\})"
+            },
+            "type": {
+              "const": "function"
+            }
+          },
+          "type": "object"
+        },
+        "type": "array",
+        "x-regex-iterator": "<\\|tool_call>(.*?)<tool_call\\|>"
+      }
+    },
+    "type": "object",
+    "x-regex": "(\\<\\|channel\\>thought\\n(?P<thinking>.*?)\\<channel\\|\\>)?(?P<tool_calls>\\<\\|tool_call\\>.*\\<tool_call\\|\\>)?(?P<content>(?:(?!\\<turn\\|\\>)(?!\\<\\|tool_response\\>).)+)?(?:\\<turn\\|\\>|\\<\\|tool_response\\>)?"
+  },
+  "soc_token": "<|channel>",
+  "sot_token": "<|turn>",
+  "stc_token": "<|tool_call>",
+  "std_token": "<|tool>",
+  "str_token": "<|tool_response>",
+  "think_token": "<|think|>",
+  "tokenizer_class": "GemmaTokenizer",
+  "unk_token": "<unk>",
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<pad>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "1": {
+      "content": "<eos>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "2": {
+      "content": "<bos>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "3": {
+      "content": "<unk>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "4": {
+      "content": "<mask>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "46": {
+      "content": "<|tool>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "47": {
+      "content": "<tool|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "48": {
+      "content": "<|tool_call>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "49": {
+      "content": "<tool_call|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "50": {
+      "content": "<|tool_response>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "51": {
+      "content": "<tool_response|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "52": {
+      "content": "<|\"|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "98": {
+      "content": "<|think|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "100": {
+      "content": "<|channel>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "101": {
+      "content": "<channel|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "105": {
+      "content": "<|turn>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "106": {
+      "content": "<turn|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "255999": {
+      "content": "<|image>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "256000": {
+      "content": "<|audio>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "258880": {
+      "content": "<|image|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "258881": {
+      "content": "<|audio|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "258882": {
+      "content": "<image|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "258883": {
+      "content": "<audio|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    "258884": {
+      "content": "<|video|>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    }
+  }
+}

unsloth_compiled_cache/AqlmLoraLinear_peft_forward.py

@@ -0,0 +1,89 @@
+"""
+2026.6.7
+2026.6.9
+5.5.0
+1.7.0
+__UNSLOTH_VERSIONING__
+"""
+
+# Unsloth auto generated code
+# Copyright 2023-present Daniel Han-Chen, Michael Han-Chen & the Unsloth team. All rights reserved.
+#
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Lesser General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public License
+# along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+
+torch_compile_options = {'epilogue_fusion': True, 'max_autotune': False, 'shape_padding': True, 'trace.enabled': False, 'triton.cudagraphs': False, 'debug': False, 'dce': True, 'memory_planning': True, 'coordinate_descent_tuning': False, 'trace.graph_diagram': False, 'compile_threads': 4, 'group_fusion': True, 'disable_progress': True, 'verbose_progress': False, 'triton.multi_kernel': 0, 'triton.use_block_ptr': False, 'triton.enable_persistent_tma_matmul': True, 'triton.autotune_at_compile_time': False, 'triton.cooperative_reductions': False, 'cuda.compile_opt_level': '-O2', 'cuda.enable_cuda_lto': True, 'combo_kernels': False, 'benchmark_combo_kernel': True, 'combo_kernel_foreach_dynamic_shapes': True}
+from torch import Tensor
+import torch
+import torch.nn as nn
+from torch.nn import functional as F
+from unsloth_zoo.temporary_patches.common import torch_compile
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+from peft.tuners.lora.aqlm import (torch)
+
+
+torch_addmm = torch.addmm
+torch_add   = torch.add
+# @torch.compile(fullgraph = False, dynamic = True, options = torch_compile_options)
+def lora_forward(result, lora_A, lora_B, dropout, x, scaling):
+    # Use result.dtype (bfloat16 from base layer) since x may have been cast to float32
+    # by _cast_input_dtype when autocast is disabled
+    target_dtype = result.dtype
+    xA = dropout(x).to(target_dtype) @ lora_A.weight.to(target_dtype).t()
+    # output = result + scaling * xA @ lora_B.weight.t()
+    shape = result.shape
+    output = torch_addmm(
+        result.view(-1, shape[-1]),
+        xA.view(-1, xA.shape[-1]),
+        lora_B.weight.to(target_dtype).t(),
+        alpha = scaling,
+        beta = 1,
+    ).view(shape)
+
+    bias = lora_B.bias
+    if bias is not None:
+        output = torch_add(
+            output,
+            bias.to(target_dtype),
+            alpha = scaling,
+        )
+    return output
+pass
+
+def unsloth_forward(self, x: torch.Tensor):
+    # note: logic differs from default Linear because merging is not supported
+    result = self.base_layer(x)
+
+    if self.disable_adapters:
+        return result
+
+    for active_adapter in self.active_adapters:
+        if active_adapter not in self.lora_A.keys():
+            continue
+        lora_A = self.lora_A[active_adapter]
+        lora_B = self.lora_B[active_adapter]
+        dropout = self.lora_dropout[active_adapter]
+        scaling = self.scaling[active_adapter]
+
+        requires_conversion = not torch.is_autocast_enabled()
+        if requires_conversion:
+            expected_dtype = result.dtype
+            x = self._cast_input_dtype(x, lora_A.weight.dtype)
+
+        output = lora_B(lora_A(dropout(x)))
+        if requires_conversion:
+            output = output.to(expected_dtype)
+        output = output * scaling
+        result += output
+    return result

unsloth_compiled_cache/AwqLoraLinear_peft_forward.py

@@ -0,0 +1,88 @@
+"""
+2026.6.7
+2026.6.9
+5.5.0
+1.7.0
+__UNSLOTH_VERSIONING__
+"""
+
+# Unsloth auto generated code
+# Copyright 2023-present Daniel Han-Chen, Michael Han-Chen & the Unsloth team. All rights reserved.
+#
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Lesser General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public License
+# along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+
+torch_compile_options = {'epilogue_fusion': True, 'max_autotune': False, 'shape_padding': True, 'trace.enabled': False, 'triton.cudagraphs': False, 'debug': False, 'dce': True, 'memory_planning': True, 'coordinate_descent_tuning': False, 'trace.graph_diagram': False, 'compile_threads': 4, 'group_fusion': True, 'disable_progress': True, 'verbose_progress': False, 'triton.multi_kernel': 0, 'triton.use_block_ptr': False, 'triton.enable_persistent_tma_matmul': True, 'triton.autotune_at_compile_time': False, 'triton.cooperative_reductions': False, 'cuda.compile_opt_level': '-O2', 'cuda.enable_cuda_lto': True, 'combo_kernels': False, 'benchmark_combo_kernel': True, 'combo_kernel_foreach_dynamic_shapes': True}
+from torch import Tensor
+import torch
+import torch.nn as nn
+from torch.nn import functional as F
+from unsloth_zoo.temporary_patches.common import torch_compile
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+from peft.tuners.lora.awq import (torch)
+
+
+torch_addmm = torch.addmm
+torch_add   = torch.add
+# @torch.compile(fullgraph = False, dynamic = True, options = torch_compile_options)
+def lora_forward(result, lora_A, lora_B, dropout, x, scaling):
+    # Use result.dtype (bfloat16 from base layer) since x may have been cast to float32
+    # by _cast_input_dtype when autocast is disabled
+    target_dtype = result.dtype
+    xA = dropout(x).to(target_dtype) @ lora_A.weight.to(target_dtype).t()
+    # output = result + scaling * xA @ lora_B.weight.t()
+    shape = result.shape
+    output = torch_addmm(
+        result.view(-1, shape[-1]),
+        xA.view(-1, xA.shape[-1]),
+        lora_B.weight.to(target_dtype).t(),
+        alpha = scaling,
+        beta = 1,
+    ).view(shape)
+
+    bias = lora_B.bias
+    if bias is not None:
+        output = torch_add(
+            output,
+            bias.to(target_dtype),
+            alpha = scaling,
+        )
+    return output
+pass
+
+def unsloth_forward(self, x: torch.Tensor):
+    result = self.quant_linear_module(x)
+
+    if self.disable_adapters:
+        return result
+
+    for active_adapter in self.active_adapters:
+        if active_adapter not in self.lora_A.keys():
+            continue
+        lora_A = self.lora_A[active_adapter]
+        lora_B = self.lora_B[active_adapter]
+        dropout = self.lora_dropout[active_adapter]
+        scaling = self.scaling[active_adapter]
+
+        requires_conversion = not torch.is_autocast_enabled()
+        if requires_conversion:
+            expected_dtype = result.dtype
+            x = self._cast_input_dtype(x, lora_A.weight.dtype)
+
+        output = lora_B(lora_A(dropout(x)))
+        if requires_conversion:
+            output = output.to(expected_dtype)
+        output = output * scaling
+        result = result + output
+    return result

unsloth_compiled_cache/BatchNorm1d.py

@@ -0,0 +1,121 @@
+"""
+2026.6.7
+2026.6.9
+5.5.0
+1.7.0
+__UNSLOTH_VERSIONING__
+"""
+
+# Unsloth auto generated code
+# Copyright 2023-present Daniel Han-Chen, Michael Han-Chen & the Unsloth team. All rights reserved.
+#
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Lesser General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public License
+# along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+
+import os
+import sys
+import torch
+import importlib.util
+import math
+if importlib.util.find_spec("unsloth_studio") is None:
+    UNSLOTH_STUDIO_ENABLED = False
+else:
+    UNSLOTH_STUDIO_ENABLED = os.environ.get("UNSLOTH_STUDIO_DISABLED", "0") == "0"
+pass
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+import math
+
+UNSLOTH_ENABLE_LOGGING = os.environ.get("UNSLOTH_ENABLE_LOGGING", "0") == "1"
+UNSLOTH_ENABLE_CCE = os.environ.get("UNSLOTH_ENABLE_CCE", "1") == "1"
+UNSLOTH_COMPILE_DISABLE = os.environ.get("UNSLOTH_COMPILE_DISABLE", "0") in ("1", "partial",)
+UNSLOTH_COMPILE_LOCATION = os.environ.get("UNSLOTH_COMPILE_LOCATION", "unsloth_compiled_cache")
+if UNSLOTH_COMPILE_LOCATION not in sys.path:
+    sys.path.insert(0, UNSLOTH_COMPILE_LOCATION)
+
+import logging
+logger_compiler = logging.getLogger(__name__)
+if UNSLOTH_ENABLE_LOGGING:
+    logger_compiler.setLevel(logging.DEBUG)
+
+global INFERENCE_RUNS
+INFERENCE_RUNS = 0
+
+try:
+    import torch._dynamo.eval_frame as torch_dynamo_eval_frame
+    torch_dynamo_eval_frame._stance.stance
+    torch_compiler_set_stance = torch.compiler.set_stance
+except:
+    torch_dynamo_eval_frame = None
+    torch_compiler_set_stance = None
+pass
+
+from unsloth_zoo import DEVICE_TYPE_TORCH, DEVICE_COUNT
+
+torch_compile_options = {'epilogue_fusion': True, 'max_autotune': False, 'shape_padding': True, 'trace.enabled': False, 'triton.cudagraphs': False, 'debug': False, 'dce': True, 'memory_planning': True, 'coordinate_descent_tuning': False, 'trace.graph_diagram': False, 'compile_threads': 4, 'group_fusion': True, 'disable_progress': True, 'verbose_progress': False, 'triton.multi_kernel': 0, 'triton.use_block_ptr': False, 'triton.enable_persistent_tma_matmul': True, 'triton.autotune_at_compile_time': False, 'triton.cooperative_reductions': False, 'cuda.compile_opt_level': '-O2', 'cuda.enable_cuda_lto': True, 'combo_kernels': False, 'benchmark_combo_kernel': True, 'combo_kernel_foreach_dynamic_shapes': True}
+from torch import Tensor
+import torch
+import torch.nn as nn
+from torch.nn import functional as F
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+from transformers.models.gemma4.modeling_gemma4 import (F, nn)
+
+def forward(self, input: Tensor) -> Tensor:
+    self._check_input_dim(input)
+
+    # exponential_average_factor is set to self.momentum
+    # (when it is available) only so that it gets updated
+    # in ONNX graph when this node is exported to ONNX.
+    if self.momentum is None:
+        exponential_average_factor = 0.0
+    else:
+        exponential_average_factor = self.momentum
+
+    if self.training and self.track_running_stats:
+        # TODO: if statement only here to tell the jit to skip emitting this when it is None
+        if self.num_batches_tracked is not None:  # type: ignore[has-type]
+            self.num_batches_tracked.add_(1)  # type: ignore[has-type]
+            if self.momentum is None:  # use cumulative moving average
+                exponential_average_factor = 1.0 / float(self.num_batches_tracked)
+            else:  # use exponential moving average
+                exponential_average_factor = self.momentum
+
+    r"""
+    Decide whether the mini-batch stats should be used for normalization rather than the buffers.
+    Mini-batch stats are used in training mode, and in eval mode when buffers are None.
+    """
+    if self.training:
+        bn_training = True
+    else:
+        bn_training = (self.running_mean is None) and (self.running_var is None)
+
+    r"""
+    Buffers are only updated if they are to be tracked and we are in training mode. Thus they only need to be
+    passed when the update should occur (i.e. in training mode when they are tracked), or when buffer stats are
+    used for normalization (i.e. in eval mode when buffers are not None).
+    """
+    return F.batch_norm(
+        input,
+        # If buffers are not to be tracked, ensure that they won't be updated
+        (
+            self.running_mean
+            if not self.training or self.track_running_stats
+            else None
+        ),
+        self.running_var if not self.training or self.track_running_stats else None,
+        self.weight,
+        self.bias,
+        bn_training,
+        exponential_average_factor,
+        self.eps,
+    ).to(input.dtype).to(input.dtype)

unsloth_compiled_cache/BatchNorm2d.py

@@ -0,0 +1,121 @@
+"""
+2026.6.7
+2026.6.9
+5.5.0
+1.7.0
+__UNSLOTH_VERSIONING__
+"""
+
+# Unsloth auto generated code
+# Copyright 2023-present Daniel Han-Chen, Michael Han-Chen & the Unsloth team. All rights reserved.
+#
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Lesser General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public License
+# along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+
+import os
+import sys
+import torch
+import importlib.util
+import math
+if importlib.util.find_spec("unsloth_studio") is None:
+    UNSLOTH_STUDIO_ENABLED = False
+else:
+    UNSLOTH_STUDIO_ENABLED = os.environ.get("UNSLOTH_STUDIO_DISABLED", "0") == "0"
+pass
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+import math
+
+UNSLOTH_ENABLE_LOGGING = os.environ.get("UNSLOTH_ENABLE_LOGGING", "0") == "1"
+UNSLOTH_ENABLE_CCE = os.environ.get("UNSLOTH_ENABLE_CCE", "1") == "1"
+UNSLOTH_COMPILE_DISABLE = os.environ.get("UNSLOTH_COMPILE_DISABLE", "0") in ("1", "partial",)
+UNSLOTH_COMPILE_LOCATION = os.environ.get("UNSLOTH_COMPILE_LOCATION", "unsloth_compiled_cache")
+if UNSLOTH_COMPILE_LOCATION not in sys.path:
+    sys.path.insert(0, UNSLOTH_COMPILE_LOCATION)
+
+import logging
+logger_compiler = logging.getLogger(__name__)
+if UNSLOTH_ENABLE_LOGGING:
+    logger_compiler.setLevel(logging.DEBUG)
+
+global INFERENCE_RUNS
+INFERENCE_RUNS = 0
+
+try:
+    import torch._dynamo.eval_frame as torch_dynamo_eval_frame
+    torch_dynamo_eval_frame._stance.stance
+    torch_compiler_set_stance = torch.compiler.set_stance
+except:
+    torch_dynamo_eval_frame = None
+    torch_compiler_set_stance = None
+pass
+
+from unsloth_zoo import DEVICE_TYPE_TORCH, DEVICE_COUNT
+
+torch_compile_options = {'epilogue_fusion': True, 'max_autotune': False, 'shape_padding': True, 'trace.enabled': False, 'triton.cudagraphs': False, 'debug': False, 'dce': True, 'memory_planning': True, 'coordinate_descent_tuning': False, 'trace.graph_diagram': False, 'compile_threads': 4, 'group_fusion': True, 'disable_progress': True, 'verbose_progress': False, 'triton.multi_kernel': 0, 'triton.use_block_ptr': False, 'triton.enable_persistent_tma_matmul': True, 'triton.autotune_at_compile_time': False, 'triton.cooperative_reductions': False, 'cuda.compile_opt_level': '-O2', 'cuda.enable_cuda_lto': True, 'combo_kernels': False, 'benchmark_combo_kernel': True, 'combo_kernel_foreach_dynamic_shapes': True}
+from torch import Tensor
+import torch
+import torch.nn as nn
+from torch.nn import functional as F
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+from transformers.models.gemma4.modeling_gemma4 import (F, nn)
+
+def forward(self, input: Tensor) -> Tensor:
+    self._check_input_dim(input)
+
+    # exponential_average_factor is set to self.momentum
+    # (when it is available) only so that it gets updated
+    # in ONNX graph when this node is exported to ONNX.
+    if self.momentum is None:
+        exponential_average_factor = 0.0
+    else:
+        exponential_average_factor = self.momentum
+
+    if self.training and self.track_running_stats:
+        # TODO: if statement only here to tell the jit to skip emitting this when it is None
+        if self.num_batches_tracked is not None:  # type: ignore[has-type]
+            self.num_batches_tracked.add_(1)  # type: ignore[has-type]
+            if self.momentum is None:  # use cumulative moving average
+                exponential_average_factor = 1.0 / float(self.num_batches_tracked)
+            else:  # use exponential moving average
+                exponential_average_factor = self.momentum
+
+    r"""
+    Decide whether the mini-batch stats should be used for normalization rather than the buffers.
+    Mini-batch stats are used in training mode, and in eval mode when buffers are None.
+    """
+    if self.training:
+        bn_training = True
+    else:
+        bn_training = (self.running_mean is None) and (self.running_var is None)
+
+    r"""
+    Buffers are only updated if they are to be tracked and we are in training mode. Thus they only need to be
+    passed when the update should occur (i.e. in training mode when they are tracked), or when buffer stats are
+    used for normalization (i.e. in eval mode when buffers are not None).
+    """
+    return F.batch_norm(
+        input,
+        # If buffers are not to be tracked, ensure that they won't be updated
+        (
+            self.running_mean
+            if not self.training or self.track_running_stats
+            else None
+        ),
+        self.running_var if not self.training or self.track_running_stats else None,
+        self.weight,
+        self.bias,
+        bn_training,
+        exponential_average_factor,
+        self.eps,
+    ).to(input.dtype).to(input.dtype)

unsloth_compiled_cache/BatchNorm3d.py

@@ -0,0 +1,121 @@
+"""
+2026.6.7
+2026.6.9
+5.5.0
+1.7.0
+__UNSLOTH_VERSIONING__
+"""
+
+# Unsloth auto generated code
+# Copyright 2023-present Daniel Han-Chen, Michael Han-Chen & the Unsloth team. All rights reserved.
+#
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Lesser General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public License
+# along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+
+import os
+import sys
+import torch
+import importlib.util
+import math
+if importlib.util.find_spec("unsloth_studio") is None:
+    UNSLOTH_STUDIO_ENABLED = False
+else:
+    UNSLOTH_STUDIO_ENABLED = os.environ.get("UNSLOTH_STUDIO_DISABLED", "0") == "0"
+pass
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+import math
+
+UNSLOTH_ENABLE_LOGGING = os.environ.get("UNSLOTH_ENABLE_LOGGING", "0") == "1"
+UNSLOTH_ENABLE_CCE = os.environ.get("UNSLOTH_ENABLE_CCE", "1") == "1"
+UNSLOTH_COMPILE_DISABLE = os.environ.get("UNSLOTH_COMPILE_DISABLE", "0") in ("1", "partial",)
+UNSLOTH_COMPILE_LOCATION = os.environ.get("UNSLOTH_COMPILE_LOCATION", "unsloth_compiled_cache")
+if UNSLOTH_COMPILE_LOCATION not in sys.path:
+    sys.path.insert(0, UNSLOTH_COMPILE_LOCATION)
+
+import logging
+logger_compiler = logging.getLogger(__name__)
+if UNSLOTH_ENABLE_LOGGING:
+    logger_compiler.setLevel(logging.DEBUG)
+
+global INFERENCE_RUNS
+INFERENCE_RUNS = 0
+
+try:
+    import torch._dynamo.eval_frame as torch_dynamo_eval_frame
+    torch_dynamo_eval_frame._stance.stance
+    torch_compiler_set_stance = torch.compiler.set_stance
+except:
+    torch_dynamo_eval_frame = None
+    torch_compiler_set_stance = None
+pass
+
+from unsloth_zoo import DEVICE_TYPE_TORCH, DEVICE_COUNT
+
+torch_compile_options = {'epilogue_fusion': True, 'max_autotune': False, 'shape_padding': True, 'trace.enabled': False, 'triton.cudagraphs': False, 'debug': False, 'dce': True, 'memory_planning': True, 'coordinate_descent_tuning': False, 'trace.graph_diagram': False, 'compile_threads': 4, 'group_fusion': True, 'disable_progress': True, 'verbose_progress': False, 'triton.multi_kernel': 0, 'triton.use_block_ptr': False, 'triton.enable_persistent_tma_matmul': True, 'triton.autotune_at_compile_time': False, 'triton.cooperative_reductions': False, 'cuda.compile_opt_level': '-O2', 'cuda.enable_cuda_lto': True, 'combo_kernels': False, 'benchmark_combo_kernel': True, 'combo_kernel_foreach_dynamic_shapes': True}
+from torch import Tensor
+import torch
+import torch.nn as nn
+from torch.nn import functional as F
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+from transformers.models.gemma4.modeling_gemma4 import (F, nn)
+
+def forward(self, input: Tensor) -> Tensor:
+    self._check_input_dim(input)
+
+    # exponential_average_factor is set to self.momentum
+    # (when it is available) only so that it gets updated
+    # in ONNX graph when this node is exported to ONNX.
+    if self.momentum is None:
+        exponential_average_factor = 0.0
+    else:
+        exponential_average_factor = self.momentum
+
+    if self.training and self.track_running_stats:
+        # TODO: if statement only here to tell the jit to skip emitting this when it is None
+        if self.num_batches_tracked is not None:  # type: ignore[has-type]
+            self.num_batches_tracked.add_(1)  # type: ignore[has-type]
+            if self.momentum is None:  # use cumulative moving average
+                exponential_average_factor = 1.0 / float(self.num_batches_tracked)
+            else:  # use exponential moving average
+                exponential_average_factor = self.momentum
+
+    r"""
+    Decide whether the mini-batch stats should be used for normalization rather than the buffers.
+    Mini-batch stats are used in training mode, and in eval mode when buffers are None.
+    """
+    if self.training:
+        bn_training = True
+    else:
+        bn_training = (self.running_mean is None) and (self.running_var is None)
+
+    r"""
+    Buffers are only updated if they are to be tracked and we are in training mode. Thus they only need to be
+    passed when the update should occur (i.e. in training mode when they are tracked), or when buffer stats are
+    used for normalization (i.e. in eval mode when buffers are not None).
+    """
+    return F.batch_norm(
+        input,
+        # If buffers are not to be tracked, ensure that they won't be updated
+        (
+            self.running_mean
+            if not self.training or self.track_running_stats
+            else None
+        ),
+        self.running_var if not self.training or self.track_running_stats else None,
+        self.weight,
+        self.bias,
+        bn_training,
+        exponential_average_factor,
+        self.eps,
+    ).to(input.dtype).to(input.dtype)

unsloth_compiled_cache/BlockDiagonalLinear_peft_forward.py

@@ -0,0 +1,75 @@
+"""
+2026.6.7
+2026.6.9
+5.5.0
+1.7.0
+__UNSLOTH_VERSIONING__
+"""
+
+# Unsloth auto generated code
+# Copyright 2023-present Daniel Han-Chen, Michael Han-Chen & the Unsloth team. All rights reserved.
+#
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Lesser General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public License
+# along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+
+torch_compile_options = {'epilogue_fusion': True, 'max_autotune': False, 'shape_padding': True, 'trace.enabled': False, 'triton.cudagraphs': False, 'debug': False, 'dce': True, 'memory_planning': True, 'coordinate_descent_tuning': False, 'trace.graph_diagram': False, 'compile_threads': 4, 'group_fusion': True, 'disable_progress': True, 'verbose_progress': False, 'triton.multi_kernel': 0, 'triton.use_block_ptr': False, 'triton.enable_persistent_tma_matmul': True, 'triton.autotune_at_compile_time': False, 'triton.cooperative_reductions': False, 'cuda.compile_opt_level': '-O2', 'cuda.enable_cuda_lto': True, 'combo_kernels': False, 'benchmark_combo_kernel': True, 'combo_kernel_foreach_dynamic_shapes': True}
+from torch import Tensor
+import torch
+import torch.nn as nn
+from torch.nn import functional as F
+from unsloth_zoo.temporary_patches.common import torch_compile
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+from peft.tuners.lora.variants import (torch)
+
+
+torch_addmm = torch.addmm
+torch_add   = torch.add
+# @torch.compile(fullgraph = False, dynamic = True, options = torch_compile_options)
+def lora_forward(result, lora_A, lora_B, dropout, x, scaling):
+    # Use result.dtype (bfloat16 from base layer) since x may have been cast to float32
+    # by _cast_input_dtype when autocast is disabled
+    target_dtype = result.dtype
+    xA = dropout(x).to(target_dtype) @ lora_A.weight.to(target_dtype).t()
+    # output = result + scaling * xA @ lora_B.weight.t()
+    shape = result.shape
+    output = torch_addmm(
+        result.view(-1, shape[-1]),
+        xA.view(-1, xA.shape[-1]),
+        lora_B.weight.to(target_dtype).t(),
+        alpha = scaling,
+        beta = 1,
+    ).view(shape)
+
+    bias = lora_B.bias
+    if bias is not None:
+        output = torch_add(
+            output,
+            bias.to(target_dtype),
+            alpha = scaling,
+        )
+    return output
+pass
+
+def unsloth_forward(self, x: torch.Tensor) -> torch.Tensor:
+    first_dims = x.shape[:-1]
+    if x.dim() != 2:
+        x = x.reshape(-1, x.shape[-1])
+    B = x.shape[0]
+    nb = self.nblocks
+    m = x.shape[-1] // nb
+    n = self.out_features // nb
+    x = x.reshape(B, nb, m)
+    w = self.weight.view(nb, n, m)
+    out = torch.einsum("bim,inm->bin", x, w)
+    return out.reshape(*first_dims, -1)

unsloth_compiled_cache/Conv1d.py

@@ -0,0 +1,78 @@
+"""
+2026.6.7
+2026.6.9
+5.5.0
+1.7.0
+__UNSLOTH_VERSIONING__
+"""
+
+# Unsloth auto generated code
+# Copyright 2023-present Daniel Han-Chen, Michael Han-Chen & the Unsloth team. All rights reserved.
+#
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Lesser General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public License
+# along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+
+import os
+import sys
+import torch
+import importlib.util
+import math
+if importlib.util.find_spec("unsloth_studio") is None:
+    UNSLOTH_STUDIO_ENABLED = False
+else:
+    UNSLOTH_STUDIO_ENABLED = os.environ.get("UNSLOTH_STUDIO_DISABLED", "0") == "0"
+pass
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+import math
+
+UNSLOTH_ENABLE_LOGGING = os.environ.get("UNSLOTH_ENABLE_LOGGING", "0") == "1"
+UNSLOTH_ENABLE_CCE = os.environ.get("UNSLOTH_ENABLE_CCE", "1") == "1"
+UNSLOTH_COMPILE_DISABLE = os.environ.get("UNSLOTH_COMPILE_DISABLE", "0") in ("1", "partial",)
+UNSLOTH_COMPILE_LOCATION = os.environ.get("UNSLOTH_COMPILE_LOCATION", "unsloth_compiled_cache")
+if UNSLOTH_COMPILE_LOCATION not in sys.path:
+    sys.path.insert(0, UNSLOTH_COMPILE_LOCATION)
+
+import logging
+logger_compiler = logging.getLogger(__name__)
+if UNSLOTH_ENABLE_LOGGING:
+    logger_compiler.setLevel(logging.DEBUG)
+
+global INFERENCE_RUNS
+INFERENCE_RUNS = 0
+
+try:
+    import torch._dynamo.eval_frame as torch_dynamo_eval_frame
+    torch_dynamo_eval_frame._stance.stance
+    torch_compiler_set_stance = torch.compiler.set_stance
+except:
+    torch_dynamo_eval_frame = None
+    torch_compiler_set_stance = None
+pass
+
+from unsloth_zoo import DEVICE_TYPE_TORCH, DEVICE_COUNT
+
+torch_compile_options = {'epilogue_fusion': True, 'max_autotune': False, 'shape_padding': True, 'trace.enabled': False, 'triton.cudagraphs': False, 'debug': False, 'dce': True, 'memory_planning': True, 'coordinate_descent_tuning': False, 'trace.graph_diagram': False, 'compile_threads': 4, 'group_fusion': True, 'disable_progress': True, 'verbose_progress': False, 'triton.multi_kernel': 0, 'triton.use_block_ptr': False, 'triton.enable_persistent_tma_matmul': True, 'triton.autotune_at_compile_time': False, 'triton.cooperative_reductions': False, 'cuda.compile_opt_level': '-O2', 'cuda.enable_cuda_lto': True, 'combo_kernels': False, 'benchmark_combo_kernel': True, 'combo_kernel_foreach_dynamic_shapes': True}
+from torch import Tensor
+import torch
+import torch.nn as nn
+from torch.nn import functional as F
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+
+
+def forward(self, input: Tensor) -> Tensor:
+    original_dtype = input.dtype
+    input = input.to(self.weight.dtype)
+    original_dtype = input.dtype
+    input = input.to(self.weight.dtype)
+    return self._conv_forward(input, self.weight, self.bias).to(original_dtype).to(original_dtype)

unsloth_compiled_cache/Conv2d.py

@@ -0,0 +1,78 @@
+"""
+2026.6.7
+2026.6.9
+5.5.0
+1.7.0
+__UNSLOTH_VERSIONING__
+"""
+
+# Unsloth auto generated code
+# Copyright 2023-present Daniel Han-Chen, Michael Han-Chen & the Unsloth team. All rights reserved.
+#
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Lesser General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public License
+# along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+
+import os
+import sys
+import torch
+import importlib.util
+import math
+if importlib.util.find_spec("unsloth_studio") is None:
+    UNSLOTH_STUDIO_ENABLED = False
+else:
+    UNSLOTH_STUDIO_ENABLED = os.environ.get("UNSLOTH_STUDIO_DISABLED", "0") == "0"
+pass
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+import math
+
+UNSLOTH_ENABLE_LOGGING = os.environ.get("UNSLOTH_ENABLE_LOGGING", "0") == "1"
+UNSLOTH_ENABLE_CCE = os.environ.get("UNSLOTH_ENABLE_CCE", "1") == "1"
+UNSLOTH_COMPILE_DISABLE = os.environ.get("UNSLOTH_COMPILE_DISABLE", "0") in ("1", "partial",)
+UNSLOTH_COMPILE_LOCATION = os.environ.get("UNSLOTH_COMPILE_LOCATION", "unsloth_compiled_cache")
+if UNSLOTH_COMPILE_LOCATION not in sys.path:
+    sys.path.insert(0, UNSLOTH_COMPILE_LOCATION)
+
+import logging
+logger_compiler = logging.getLogger(__name__)
+if UNSLOTH_ENABLE_LOGGING:
+    logger_compiler.setLevel(logging.DEBUG)
+
+global INFERENCE_RUNS
+INFERENCE_RUNS = 0
+
+try:
+    import torch._dynamo.eval_frame as torch_dynamo_eval_frame
+    torch_dynamo_eval_frame._stance.stance
+    torch_compiler_set_stance = torch.compiler.set_stance
+except:
+    torch_dynamo_eval_frame = None
+    torch_compiler_set_stance = None
+pass
+
+from unsloth_zoo import DEVICE_TYPE_TORCH, DEVICE_COUNT
+
+torch_compile_options = {'epilogue_fusion': True, 'max_autotune': False, 'shape_padding': True, 'trace.enabled': False, 'triton.cudagraphs': False, 'debug': False, 'dce': True, 'memory_planning': True, 'coordinate_descent_tuning': False, 'trace.graph_diagram': False, 'compile_threads': 4, 'group_fusion': True, 'disable_progress': True, 'verbose_progress': False, 'triton.multi_kernel': 0, 'triton.use_block_ptr': False, 'triton.enable_persistent_tma_matmul': True, 'triton.autotune_at_compile_time': False, 'triton.cooperative_reductions': False, 'cuda.compile_opt_level': '-O2', 'cuda.enable_cuda_lto': True, 'combo_kernels': False, 'benchmark_combo_kernel': True, 'combo_kernel_foreach_dynamic_shapes': True}
+from torch import Tensor
+import torch
+import torch.nn as nn
+from torch.nn import functional as F
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+
+
+def forward(self, input: Tensor) -> Tensor:
+    original_dtype = input.dtype
+    input = input.to(self.weight.dtype)
+    original_dtype = input.dtype
+    input = input.to(self.weight.dtype)
+    return self._conv_forward(input, self.weight, self.bias).to(original_dtype).to(original_dtype)

unsloth_compiled_cache/Conv3d.py

@@ -0,0 +1,78 @@
+"""
+2026.6.7
+2026.6.9
+5.5.0
+1.7.0
+__UNSLOTH_VERSIONING__
+"""
+
+# Unsloth auto generated code
+# Copyright 2023-present Daniel Han-Chen, Michael Han-Chen & the Unsloth team. All rights reserved.
+#
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Lesser General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public License
+# along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+
+import os
+import sys
+import torch
+import importlib.util
+import math
+if importlib.util.find_spec("unsloth_studio") is None:
+    UNSLOTH_STUDIO_ENABLED = False
+else:
+    UNSLOTH_STUDIO_ENABLED = os.environ.get("UNSLOTH_STUDIO_DISABLED", "0") == "0"
+pass
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+import math
+
+UNSLOTH_ENABLE_LOGGING = os.environ.get("UNSLOTH_ENABLE_LOGGING", "0") == "1"
+UNSLOTH_ENABLE_CCE = os.environ.get("UNSLOTH_ENABLE_CCE", "1") == "1"
+UNSLOTH_COMPILE_DISABLE = os.environ.get("UNSLOTH_COMPILE_DISABLE", "0") in ("1", "partial",)
+UNSLOTH_COMPILE_LOCATION = os.environ.get("UNSLOTH_COMPILE_LOCATION", "unsloth_compiled_cache")
+if UNSLOTH_COMPILE_LOCATION not in sys.path:
+    sys.path.insert(0, UNSLOTH_COMPILE_LOCATION)
+
+import logging
+logger_compiler = logging.getLogger(__name__)
+if UNSLOTH_ENABLE_LOGGING:
+    logger_compiler.setLevel(logging.DEBUG)
+
+global INFERENCE_RUNS
+INFERENCE_RUNS = 0
+
+try:
+    import torch._dynamo.eval_frame as torch_dynamo_eval_frame
+    torch_dynamo_eval_frame._stance.stance
+    torch_compiler_set_stance = torch.compiler.set_stance
+except:
+    torch_dynamo_eval_frame = None
+    torch_compiler_set_stance = None
+pass
+
+from unsloth_zoo import DEVICE_TYPE_TORCH, DEVICE_COUNT
+
+torch_compile_options = {'epilogue_fusion': True, 'max_autotune': False, 'shape_padding': True, 'trace.enabled': False, 'triton.cudagraphs': False, 'debug': False, 'dce': True, 'memory_planning': True, 'coordinate_descent_tuning': False, 'trace.graph_diagram': False, 'compile_threads': 4, 'group_fusion': True, 'disable_progress': True, 'verbose_progress': False, 'triton.multi_kernel': 0, 'triton.use_block_ptr': False, 'triton.enable_persistent_tma_matmul': True, 'triton.autotune_at_compile_time': False, 'triton.cooperative_reductions': False, 'cuda.compile_opt_level': '-O2', 'cuda.enable_cuda_lto': True, 'combo_kernels': False, 'benchmark_combo_kernel': True, 'combo_kernel_foreach_dynamic_shapes': True}
+from torch import Tensor
+import torch
+import torch.nn as nn
+from torch.nn import functional as F
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+
+
+def forward(self, input: Tensor) -> Tensor:
+    original_dtype = input.dtype
+    input = input.to(self.weight.dtype)
+    original_dtype = input.dtype
+    input = input.to(self.weight.dtype)
+    return self._conv_forward(input, self.weight, self.bias).to(original_dtype).to(original_dtype)

unsloth_compiled_cache/ConvTranspose1d.py

@@ -0,0 +1,105 @@
+"""
+2026.6.7
+2026.6.9
+5.5.0
+1.7.0
+__UNSLOTH_VERSIONING__
+"""
+
+# Unsloth auto generated code
+# Copyright 2023-present Daniel Han-Chen, Michael Han-Chen & the Unsloth team. All rights reserved.
+#
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Lesser General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public License
+# along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+
+import os
+import sys
+import torch
+import importlib.util
+import math
+if importlib.util.find_spec("unsloth_studio") is None:
+    UNSLOTH_STUDIO_ENABLED = False
+else:
+    UNSLOTH_STUDIO_ENABLED = os.environ.get("UNSLOTH_STUDIO_DISABLED", "0") == "0"
+pass
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+import math
+
+UNSLOTH_ENABLE_LOGGING = os.environ.get("UNSLOTH_ENABLE_LOGGING", "0") == "1"
+UNSLOTH_ENABLE_CCE = os.environ.get("UNSLOTH_ENABLE_CCE", "1") == "1"
+UNSLOTH_COMPILE_DISABLE = os.environ.get("UNSLOTH_COMPILE_DISABLE", "0") in ("1", "partial",)
+UNSLOTH_COMPILE_LOCATION = os.environ.get("UNSLOTH_COMPILE_LOCATION", "unsloth_compiled_cache")
+if UNSLOTH_COMPILE_LOCATION not in sys.path:
+    sys.path.insert(0, UNSLOTH_COMPILE_LOCATION)
+
+import logging
+logger_compiler = logging.getLogger(__name__)
+if UNSLOTH_ENABLE_LOGGING:
+    logger_compiler.setLevel(logging.DEBUG)
+
+global INFERENCE_RUNS
+INFERENCE_RUNS = 0
+
+try:
+    import torch._dynamo.eval_frame as torch_dynamo_eval_frame
+    torch_dynamo_eval_frame._stance.stance
+    torch_compiler_set_stance = torch.compiler.set_stance
+except:
+    torch_dynamo_eval_frame = None
+    torch_compiler_set_stance = None
+pass
+
+from unsloth_zoo import DEVICE_TYPE_TORCH, DEVICE_COUNT
+
+torch_compile_options = {'epilogue_fusion': True, 'max_autotune': False, 'shape_padding': True, 'trace.enabled': False, 'triton.cudagraphs': False, 'debug': False, 'dce': True, 'memory_planning': True, 'coordinate_descent_tuning': False, 'trace.graph_diagram': False, 'compile_threads': 4, 'group_fusion': True, 'disable_progress': True, 'verbose_progress': False, 'triton.multi_kernel': 0, 'triton.use_block_ptr': False, 'triton.enable_persistent_tma_matmul': True, 'triton.autotune_at_compile_time': False, 'triton.cooperative_reductions': False, 'cuda.compile_opt_level': '-O2', 'cuda.enable_cuda_lto': True, 'combo_kernels': False, 'benchmark_combo_kernel': True, 'combo_kernel_foreach_dynamic_shapes': True}
+from torch import Tensor
+import torch
+import torch.nn as nn
+from torch.nn import functional as F
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+from transformers.models.gemma4.modeling_gemma4 import (F, nn)
+
+def forward(self, input: Tensor, output_size: list[int] | None = None) -> Tensor:
+    original_dtype = input.dtype
+    input = input.to(self.weight.dtype)
+    original_dtype = input.dtype
+    input = input.to(self.weight.dtype)
+    if self.padding_mode != "zeros":
+        raise ValueError(
+            "Only `zeros` padding mode is supported for ConvTranspose1d"
+        )
+
+    assert isinstance(self.padding, tuple)
+    # One cannot replace List by Tuple or Sequence in "_output_padding" because
+    # TorchScript does not support `Sequence[T]` or `Tuple[T, ...]`.
+    num_spatial_dims = 1
+    output_padding = self._output_padding(
+        input,
+        output_size,
+        self.stride,  # type: ignore[arg-type]
+        self.padding,  # type: ignore[arg-type]
+        self.kernel_size,  # type: ignore[arg-type]
+        num_spatial_dims,
+        self.dilation,  # type: ignore[arg-type]
+    )
+    return F.conv_transpose1d(
+        input,
+        self.weight,
+        self.bias,
+        self.stride,
+        self.padding,
+        output_padding,
+        self.groups,
+        self.dilation,
+    ).to(original_dtype).to(original_dtype)

unsloth_compiled_cache/ConvTranspose2d.py

@@ -0,0 +1,114 @@
+"""
+2026.6.7
+2026.6.9
+5.5.0
+1.7.0
+__UNSLOTH_VERSIONING__
+"""
+
+# Unsloth auto generated code
+# Copyright 2023-present Daniel Han-Chen, Michael Han-Chen & the Unsloth team. All rights reserved.
+#
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Lesser General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public License
+# along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+
+import os
+import sys
+import torch
+import importlib.util
+import math
+if importlib.util.find_spec("unsloth_studio") is None:
+    UNSLOTH_STUDIO_ENABLED = False
+else:
+    UNSLOTH_STUDIO_ENABLED = os.environ.get("UNSLOTH_STUDIO_DISABLED", "0") == "0"
+pass
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+import math
+
+UNSLOTH_ENABLE_LOGGING = os.environ.get("UNSLOTH_ENABLE_LOGGING", "0") == "1"
+UNSLOTH_ENABLE_CCE = os.environ.get("UNSLOTH_ENABLE_CCE", "1") == "1"
+UNSLOTH_COMPILE_DISABLE = os.environ.get("UNSLOTH_COMPILE_DISABLE", "0") in ("1", "partial",)
+UNSLOTH_COMPILE_LOCATION = os.environ.get("UNSLOTH_COMPILE_LOCATION", "unsloth_compiled_cache")
+if UNSLOTH_COMPILE_LOCATION not in sys.path:
+    sys.path.insert(0, UNSLOTH_COMPILE_LOCATION)
+
+import logging
+logger_compiler = logging.getLogger(__name__)
+if UNSLOTH_ENABLE_LOGGING:
+    logger_compiler.setLevel(logging.DEBUG)
+
+global INFERENCE_RUNS
+INFERENCE_RUNS = 0
+
+try:
+    import torch._dynamo.eval_frame as torch_dynamo_eval_frame
+    torch_dynamo_eval_frame._stance.stance
+    torch_compiler_set_stance = torch.compiler.set_stance
+except:
+    torch_dynamo_eval_frame = None
+    torch_compiler_set_stance = None
+pass
+
+from unsloth_zoo import DEVICE_TYPE_TORCH, DEVICE_COUNT
+
+torch_compile_options = {'epilogue_fusion': True, 'max_autotune': False, 'shape_padding': True, 'trace.enabled': False, 'triton.cudagraphs': False, 'debug': False, 'dce': True, 'memory_planning': True, 'coordinate_descent_tuning': False, 'trace.graph_diagram': False, 'compile_threads': 4, 'group_fusion': True, 'disable_progress': True, 'verbose_progress': False, 'triton.multi_kernel': 0, 'triton.use_block_ptr': False, 'triton.enable_persistent_tma_matmul': True, 'triton.autotune_at_compile_time': False, 'triton.cooperative_reductions': False, 'cuda.compile_opt_level': '-O2', 'cuda.enable_cuda_lto': True, 'combo_kernels': False, 'benchmark_combo_kernel': True, 'combo_kernel_foreach_dynamic_shapes': True}
+from torch import Tensor
+import torch
+import torch.nn as nn
+from torch.nn import functional as F
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+from transformers.models.gemma4.modeling_gemma4 import (F, nn)
+
+def forward(self, input: Tensor, output_size: list[int] | None = None) -> Tensor:
+    original_dtype = input.dtype
+    input = input.to(self.weight.dtype)
+    original_dtype = input.dtype
+    input = input.to(self.weight.dtype)
+    """
+    Performs the forward pass.
+
+    Attributes:
+        input (Tensor): The input tensor.
+        output_size (list[int], optional): A list of integers representing
+            the size of the output tensor. Default is None.
+    """
+    if self.padding_mode != "zeros":
+        raise ValueError(
+            "Only `zeros` padding mode is supported for ConvTranspose2d"
+        )
+
+    assert isinstance(self.padding, tuple)
+    # One cannot replace List by Tuple or Sequence in "_output_padding" because
+    # TorchScript does not support `Sequence[T]` or `Tuple[T, ...]`.
+    num_spatial_dims = 2
+    output_padding = self._output_padding(
+        input,
+        output_size,
+        self.stride,  # type: ignore[arg-type]
+        self.padding,  # type: ignore[arg-type]
+        self.kernel_size,  # type: ignore[arg-type]
+        num_spatial_dims,
+        self.dilation,  # type: ignore[arg-type]
+    )
+
+    return F.conv_transpose2d(
+        input,
+        self.weight,
+        self.bias,
+        self.stride,
+        self.padding,
+        output_padding,
+        self.groups,
+        self.dilation,
+    ).to(original_dtype).to(original_dtype)

unsloth_compiled_cache/ConvTranspose3d.py

@@ -0,0 +1,106 @@
+"""
+2026.6.7
+2026.6.9
+5.5.0
+1.7.0
+__UNSLOTH_VERSIONING__
+"""
+
+# Unsloth auto generated code
+# Copyright 2023-present Daniel Han-Chen, Michael Han-Chen & the Unsloth team. All rights reserved.
+#
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Lesser General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public License
+# along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+
+import os
+import sys
+import torch
+import importlib.util
+import math
+if importlib.util.find_spec("unsloth_studio") is None:
+    UNSLOTH_STUDIO_ENABLED = False
+else:
+    UNSLOTH_STUDIO_ENABLED = os.environ.get("UNSLOTH_STUDIO_DISABLED", "0") == "0"
+pass
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+import math
+
+UNSLOTH_ENABLE_LOGGING = os.environ.get("UNSLOTH_ENABLE_LOGGING", "0") == "1"
+UNSLOTH_ENABLE_CCE = os.environ.get("UNSLOTH_ENABLE_CCE", "1") == "1"
+UNSLOTH_COMPILE_DISABLE = os.environ.get("UNSLOTH_COMPILE_DISABLE", "0") in ("1", "partial",)
+UNSLOTH_COMPILE_LOCATION = os.environ.get("UNSLOTH_COMPILE_LOCATION", "unsloth_compiled_cache")
+if UNSLOTH_COMPILE_LOCATION not in sys.path:
+    sys.path.insert(0, UNSLOTH_COMPILE_LOCATION)
+
+import logging
+logger_compiler = logging.getLogger(__name__)
+if UNSLOTH_ENABLE_LOGGING:
+    logger_compiler.setLevel(logging.DEBUG)
+
+global INFERENCE_RUNS
+INFERENCE_RUNS = 0
+
+try:
+    import torch._dynamo.eval_frame as torch_dynamo_eval_frame
+    torch_dynamo_eval_frame._stance.stance
+    torch_compiler_set_stance = torch.compiler.set_stance
+except:
+    torch_dynamo_eval_frame = None
+    torch_compiler_set_stance = None
+pass
+
+from unsloth_zoo import DEVICE_TYPE_TORCH, DEVICE_COUNT
+
+torch_compile_options = {'epilogue_fusion': True, 'max_autotune': False, 'shape_padding': True, 'trace.enabled': False, 'triton.cudagraphs': False, 'debug': False, 'dce': True, 'memory_planning': True, 'coordinate_descent_tuning': False, 'trace.graph_diagram': False, 'compile_threads': 4, 'group_fusion': True, 'disable_progress': True, 'verbose_progress': False, 'triton.multi_kernel': 0, 'triton.use_block_ptr': False, 'triton.enable_persistent_tma_matmul': True, 'triton.autotune_at_compile_time': False, 'triton.cooperative_reductions': False, 'cuda.compile_opt_level': '-O2', 'cuda.enable_cuda_lto': True, 'combo_kernels': False, 'benchmark_combo_kernel': True, 'combo_kernel_foreach_dynamic_shapes': True}
+from torch import Tensor
+import torch
+import torch.nn as nn
+from torch.nn import functional as F
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+from transformers.models.gemma4.modeling_gemma4 import (F, nn)
+
+def forward(self, input: Tensor, output_size: list[int] | None = None) -> Tensor:
+    original_dtype = input.dtype
+    input = input.to(self.weight.dtype)
+    original_dtype = input.dtype
+    input = input.to(self.weight.dtype)
+    if self.padding_mode != "zeros":
+        raise ValueError(
+            "Only `zeros` padding mode is supported for ConvTranspose3d"
+        )
+
+    assert isinstance(self.padding, tuple)
+    # One cannot replace List by Tuple or Sequence in "_output_padding" because
+    # TorchScript does not support `Sequence[T]` or `Tuple[T, ...]`.
+    num_spatial_dims = 3
+    output_padding = self._output_padding(
+        input,
+        output_size,
+        self.stride,  # type: ignore[arg-type]
+        self.padding,  # type: ignore[arg-type]
+        self.kernel_size,  # type: ignore[arg-type]
+        num_spatial_dims,
+        self.dilation,  # type: ignore[arg-type]
+    )
+
+    return F.conv_transpose3d(
+        input,
+        self.weight,
+        self.bias,
+        self.stride,
+        self.padding,
+        output_padding,
+        self.groups,
+        self.dilation,
+    ).to(original_dtype).to(original_dtype)

unsloth_compiled_cache/GPTQLoraLinear_peft_forward.py

@@ -0,0 +1,96 @@
+"""
+2026.6.7
+2026.6.9
+5.5.0
+1.7.0
+__UNSLOTH_VERSIONING__
+"""
+
+# Unsloth auto generated code
+# Copyright 2023-present Daniel Han-Chen, Michael Han-Chen & the Unsloth team. All rights reserved.
+#
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Lesser General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public License
+# along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+
+torch_compile_options = {'epilogue_fusion': True, 'max_autotune': False, 'shape_padding': True, 'trace.enabled': False, 'triton.cudagraphs': False, 'debug': False, 'dce': True, 'memory_planning': True, 'coordinate_descent_tuning': False, 'trace.graph_diagram': False, 'compile_threads': 4, 'group_fusion': True, 'disable_progress': True, 'verbose_progress': False, 'triton.multi_kernel': 0, 'triton.use_block_ptr': False, 'triton.enable_persistent_tma_matmul': True, 'triton.autotune_at_compile_time': False, 'triton.cooperative_reductions': False, 'cuda.compile_opt_level': '-O2', 'cuda.enable_cuda_lto': True, 'combo_kernels': False, 'benchmark_combo_kernel': True, 'combo_kernel_foreach_dynamic_shapes': True}
+from torch import Tensor
+import torch
+import torch.nn as nn
+from torch.nn import functional as F
+from unsloth_zoo.temporary_patches.common import torch_compile
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+from peft.tuners.lora.gptq import (torch)
+
+
+torch_addmm = torch.addmm
+torch_add   = torch.add
+# @torch.compile(fullgraph = False, dynamic = True, options = torch_compile_options)
+def lora_forward(result, lora_A, lora_B, dropout, x, scaling):
+    # Use result.dtype (bfloat16 from base layer) since x may have been cast to float32
+    # by _cast_input_dtype when autocast is disabled
+    target_dtype = result.dtype
+    xA = dropout(x).to(target_dtype) @ lora_A.weight.to(target_dtype).t()
+    # output = result + scaling * xA @ lora_B.weight.t()
+    shape = result.shape
+    output = torch_addmm(
+        result.view(-1, shape[-1]),
+        xA.view(-1, xA.shape[-1]),
+        lora_B.weight.to(target_dtype).t(),
+        alpha = scaling,
+        beta = 1,
+    ).view(shape)
+
+    bias = lora_B.bias
+    if bias is not None:
+        output = torch_add(
+            output,
+            bias.to(target_dtype),
+            alpha = scaling,
+        )
+    return output
+pass
+
+def unsloth_forward(self, x: torch.Tensor):
+    # note: logic differs from default Linear because merging is not supported
+    result = self.quant_linear_module(x)
+
+    if self.disable_adapters:
+        return result
+
+    lora_A_keys = self.lora_A.keys()
+
+    for active_adapter in self.active_adapters:
+        if active_adapter not in lora_A_keys:
+            continue
+        torch_result_dtype = result.dtype
+
+        lora_A = self.lora_A[active_adapter]
+        lora_B = self.lora_B[active_adapter]
+        dropout = self.lora_dropout[active_adapter]
+        scaling = self.scaling[active_adapter]
+
+        if not torch.is_autocast_enabled(): result, x = result.to(lora_A.weight.dtype), x.to(lora_A.weight.dtype)
+
+        if active_adapter not in self.lora_variant:  # vanilla LoRA
+            return lora_forward(result, lora_A, lora_B, dropout, x, scaling).to(torch_result_dtype)
+        else:
+            result = self.lora_variant[active_adapter].forward(
+                self,
+                active_adapter=active_adapter,
+                x=x,
+                result=result,
+            )
+
+        result = result.to(torch_result_dtype)
+    return result

unsloth_compiled_cache/GroupNorm.py

@@ -0,0 +1,74 @@
+"""
+2026.6.7
+2026.6.9
+5.5.0
+1.7.0
+__UNSLOTH_VERSIONING__
+"""
+
+# Unsloth auto generated code
+# Copyright 2023-present Daniel Han-Chen, Michael Han-Chen & the Unsloth team. All rights reserved.
+#
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Lesser General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public License
+# along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+
+import os
+import sys
+import torch
+import importlib.util
+import math
+if importlib.util.find_spec("unsloth_studio") is None:
+    UNSLOTH_STUDIO_ENABLED = False
+else:
+    UNSLOTH_STUDIO_ENABLED = os.environ.get("UNSLOTH_STUDIO_DISABLED", "0") == "0"
+pass
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+import math
+
+UNSLOTH_ENABLE_LOGGING = os.environ.get("UNSLOTH_ENABLE_LOGGING", "0") == "1"
+UNSLOTH_ENABLE_CCE = os.environ.get("UNSLOTH_ENABLE_CCE", "1") == "1"
+UNSLOTH_COMPILE_DISABLE = os.environ.get("UNSLOTH_COMPILE_DISABLE", "0") in ("1", "partial",)
+UNSLOTH_COMPILE_LOCATION = os.environ.get("UNSLOTH_COMPILE_LOCATION", "unsloth_compiled_cache")
+if UNSLOTH_COMPILE_LOCATION not in sys.path:
+    sys.path.insert(0, UNSLOTH_COMPILE_LOCATION)
+
+import logging
+logger_compiler = logging.getLogger(__name__)
+if UNSLOTH_ENABLE_LOGGING:
+    logger_compiler.setLevel(logging.DEBUG)
+
+global INFERENCE_RUNS
+INFERENCE_RUNS = 0
+
+try:
+    import torch._dynamo.eval_frame as torch_dynamo_eval_frame
+    torch_dynamo_eval_frame._stance.stance
+    torch_compiler_set_stance = torch.compiler.set_stance
+except:
+    torch_dynamo_eval_frame = None
+    torch_compiler_set_stance = None
+pass
+
+from unsloth_zoo import DEVICE_TYPE_TORCH, DEVICE_COUNT
+
+torch_compile_options = {'epilogue_fusion': True, 'max_autotune': False, 'shape_padding': True, 'trace.enabled': False, 'triton.cudagraphs': False, 'debug': False, 'dce': True, 'memory_planning': True, 'coordinate_descent_tuning': False, 'trace.graph_diagram': False, 'compile_threads': 4, 'group_fusion': True, 'disable_progress': True, 'verbose_progress': False, 'triton.multi_kernel': 0, 'triton.use_block_ptr': False, 'triton.enable_persistent_tma_matmul': True, 'triton.autotune_at_compile_time': False, 'triton.cooperative_reductions': False, 'cuda.compile_opt_level': '-O2', 'cuda.enable_cuda_lto': True, 'combo_kernels': False, 'benchmark_combo_kernel': True, 'combo_kernel_foreach_dynamic_shapes': True}
+from torch import Tensor
+import torch
+import torch.nn as nn
+from torch.nn import functional as F
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+from transformers.models.gemma4.modeling_gemma4 import (F)
+
+def forward(self, input: Tensor) -> Tensor:
+    return F.group_norm(input, self.num_groups, self.weight, self.bias, self.eps).to(input.dtype).to(input.dtype)

unsloth_compiled_cache/LayerNorm.py

@@ -0,0 +1,76 @@
+"""
+2026.6.7
+2026.6.9
+5.5.0
+1.7.0
+__UNSLOTH_VERSIONING__
+"""
+
+# Unsloth auto generated code
+# Copyright 2023-present Daniel Han-Chen, Michael Han-Chen & the Unsloth team. All rights reserved.
+#
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Lesser General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public License
+# along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+
+import os
+import sys
+import torch
+import importlib.util
+import math
+if importlib.util.find_spec("unsloth_studio") is None:
+    UNSLOTH_STUDIO_ENABLED = False
+else:
+    UNSLOTH_STUDIO_ENABLED = os.environ.get("UNSLOTH_STUDIO_DISABLED", "0") == "0"
+pass
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+import math
+
+UNSLOTH_ENABLE_LOGGING = os.environ.get("UNSLOTH_ENABLE_LOGGING", "0") == "1"
+UNSLOTH_ENABLE_CCE = os.environ.get("UNSLOTH_ENABLE_CCE", "1") == "1"
+UNSLOTH_COMPILE_DISABLE = os.environ.get("UNSLOTH_COMPILE_DISABLE", "0") in ("1", "partial",)
+UNSLOTH_COMPILE_LOCATION = os.environ.get("UNSLOTH_COMPILE_LOCATION", "unsloth_compiled_cache")
+if UNSLOTH_COMPILE_LOCATION not in sys.path:
+    sys.path.insert(0, UNSLOTH_COMPILE_LOCATION)
+
+import logging
+logger_compiler = logging.getLogger(__name__)
+if UNSLOTH_ENABLE_LOGGING:
+    logger_compiler.setLevel(logging.DEBUG)
+
+global INFERENCE_RUNS
+INFERENCE_RUNS = 0
+
+try:
+    import torch._dynamo.eval_frame as torch_dynamo_eval_frame
+    torch_dynamo_eval_frame._stance.stance
+    torch_compiler_set_stance = torch.compiler.set_stance
+except:
+    torch_dynamo_eval_frame = None
+    torch_compiler_set_stance = None
+pass
+
+from unsloth_zoo import DEVICE_TYPE_TORCH, DEVICE_COUNT
+
+torch_compile_options = {'epilogue_fusion': True, 'max_autotune': False, 'shape_padding': True, 'trace.enabled': False, 'triton.cudagraphs': False, 'debug': False, 'dce': True, 'memory_planning': True, 'coordinate_descent_tuning': False, 'trace.graph_diagram': False, 'compile_threads': 4, 'group_fusion': True, 'disable_progress': True, 'verbose_progress': False, 'triton.multi_kernel': 0, 'triton.use_block_ptr': False, 'triton.enable_persistent_tma_matmul': True, 'triton.autotune_at_compile_time': False, 'triton.cooperative_reductions': False, 'cuda.compile_opt_level': '-O2', 'cuda.enable_cuda_lto': True, 'combo_kernels': False, 'benchmark_combo_kernel': True, 'combo_kernel_foreach_dynamic_shapes': True}
+from torch import Tensor
+import torch
+import torch.nn as nn
+from torch.nn import functional as F
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+from transformers.models.gemma4.modeling_gemma4 import (F)
+
+def forward(self, input: Tensor) -> Tensor:
+    return F.layer_norm(
+        input, self.normalized_shape, self.weight, self.bias, self.eps
+    ).to(input.dtype).to(input.dtype)

unsloth_compiled_cache/Linear4bit_peft_forward.py

@@ -0,0 +1,126 @@
+"""
+2026.6.7
+2026.6.9
+5.5.0
+1.7.0
+__UNSLOTH_VERSIONING__
+"""
+
+# Unsloth auto generated code
+# Copyright 2023-present Daniel Han-Chen, Michael Han-Chen & the Unsloth team. All rights reserved.
+#
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Lesser General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public License
+# along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+
+try:
+    from peft.tuners.lora.layer import VARIANT_KWARG_KEYS
+except ImportError:
+    VARIANT_KWARG_KEYS = ['alora_offsets']
+torch_compile_options = {'epilogue_fusion': True, 'max_autotune': False, 'shape_padding': True, 'trace.enabled': False, 'triton.cudagraphs': False, 'debug': False, 'dce': True, 'memory_planning': True, 'coordinate_descent_tuning': False, 'trace.graph_diagram': False, 'compile_threads': 4, 'group_fusion': True, 'disable_progress': True, 'verbose_progress': False, 'triton.multi_kernel': 0, 'triton.use_block_ptr': False, 'triton.enable_persistent_tma_matmul': True, 'triton.autotune_at_compile_time': False, 'triton.cooperative_reductions': False, 'cuda.compile_opt_level': '-O2', 'cuda.enable_cuda_lto': True, 'combo_kernels': False, 'benchmark_combo_kernel': True, 'combo_kernel_foreach_dynamic_shapes': True}
+from torch import Tensor
+import torch
+import torch.nn as nn
+from torch.nn import functional as F
+from unsloth_zoo.temporary_patches.common import torch_compile
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+from peft.tuners.lora.bnb import (VARIANT_KWARG_KEYS, torch)
+
+
+torch_addmm = torch.addmm
+torch_add   = torch.add
+# @torch.compile(fullgraph = False, dynamic = True, options = torch_compile_options)
+def lora_forward(result, lora_A, lora_B, dropout, x, scaling):
+    # Use result.dtype (bfloat16 from base layer) since x may have been cast to float32
+    # by _cast_input_dtype when autocast is disabled
+    target_dtype = result.dtype
+    xA = dropout(x).to(target_dtype) @ lora_A.weight.to(target_dtype).t()
+    # output = result + scaling * xA @ lora_B.weight.t()
+    shape = result.shape
+    output = torch_addmm(
+        result.view(-1, shape[-1]),
+        xA.view(-1, xA.shape[-1]),
+        lora_B.weight.to(target_dtype).t(),
+        alpha = scaling,
+        beta = 1,
+    ).view(shape)
+
+    bias = lora_B.bias
+    if bias is not None:
+        output = torch_add(
+            output,
+            bias.to(target_dtype),
+            alpha = scaling,
+        )
+    return output
+pass
+
+def unsloth_forward(self, x: torch.Tensor, *args, **kwargs) -> torch.Tensor:
+    
+    adapter_names = kwargs.pop("adapter_names", None)
+    variant_kwargs = {k: kwargs.pop(k, None) for k in VARIANT_KWARG_KEYS}  # don't pass these to base_layer
+
+    if self.disable_adapters:
+        if self.merged:
+            self.unmerge()
+        if not torch.is_autocast_enabled() and hasattr(self.base_layer, 'weight') and self.base_layer.weight is not None and not hasattr(self.base_layer.weight, 'quant_state') and x.dtype != self.base_layer.weight.dtype:
+            x = x.to(self.base_layer.weight.dtype)
+        result = self.base_layer(x, *args, **kwargs)
+    elif adapter_names is not None:
+        result = self._mixed_batch_forward(x, *args, adapter_names=adapter_names, **variant_kwargs, **kwargs)
+    elif self.merged:
+        if not torch.is_autocast_enabled() and hasattr(self.base_layer, 'weight') and self.base_layer.weight is not None and not hasattr(self.base_layer.weight, 'quant_state') and x.dtype != self.base_layer.weight.dtype:
+            x = x.to(self.base_layer.weight.dtype)
+        result = self.base_layer(x, *args, **kwargs)
+    else:
+        if not torch.is_autocast_enabled() and hasattr(self.base_layer, 'weight') and self.base_layer.weight is not None and not hasattr(self.base_layer.weight, 'quant_state') and x.dtype != self.base_layer.weight.dtype:
+            x = x.to(self.base_layer.weight.dtype)
+        result = self.base_layer(x, *args, **kwargs)
+        # As per Tim Dettmers, for 4bit, we need to defensively clone here.
+        # The reason is that in some cases, an error can occur that backprop
+        # does not work on a manipulated view. This issue may be solved with
+        # newer PyTorch versions but this would need extensive testing to be
+        # sure.
+        
+
+        for active_adapter in self.active_adapters:
+            if active_adapter not in self.lora_A.keys():
+                continue
+            lora_A = self.lora_A[active_adapter]
+            lora_B = self.lora_B[active_adapter]
+            dropout = self.lora_dropout[active_adapter]
+            scaling = self.scaling[active_adapter]
+
+            requires_conversion = not torch.is_autocast_enabled()
+            if requires_conversion:
+                expected_dtype = result.dtype
+                x = self._cast_input_dtype(x, lora_A.weight.dtype)
+
+            if active_adapter not in self.lora_variant:  # vanilla LoRA
+                return lora_forward(result, lora_A, lora_B, dropout, x, scaling).to(result.dtype)
+                if requires_conversion:
+                    output = output.to(expected_dtype)
+                result = result + output
+            else:
+                result = self.lora_variant[active_adapter].forward(
+                    self,
+                    active_adapter=active_adapter,
+                    x=x,
+                    result=result,
+                    **variant_kwargs,
+                    **kwargs,
+                )
+                if requires_conversion:
+                    result = result.to(expected_dtype)
+
+    return result

unsloth_compiled_cache/Linear8bitLt_peft_forward.py

@@ -0,0 +1,118 @@
+"""
+2026.6.7
+2026.6.9
+5.5.0
+1.7.0
+__UNSLOTH_VERSIONING__
+"""
+
+# Unsloth auto generated code
+# Copyright 2023-present Daniel Han-Chen, Michael Han-Chen & the Unsloth team. All rights reserved.
+#
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Lesser General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public License
+# along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+
+try:
+    from peft.tuners.lora.layer import VARIANT_KWARG_KEYS
+except ImportError:
+    VARIANT_KWARG_KEYS = ['alora_offsets']
+torch_compile_options = {'epilogue_fusion': True, 'max_autotune': False, 'shape_padding': True, 'trace.enabled': False, 'triton.cudagraphs': False, 'debug': False, 'dce': True, 'memory_planning': True, 'coordinate_descent_tuning': False, 'trace.graph_diagram': False, 'compile_threads': 4, 'group_fusion': True, 'disable_progress': True, 'verbose_progress': False, 'triton.multi_kernel': 0, 'triton.use_block_ptr': False, 'triton.enable_persistent_tma_matmul': True, 'triton.autotune_at_compile_time': False, 'triton.cooperative_reductions': False, 'cuda.compile_opt_level': '-O2', 'cuda.enable_cuda_lto': True, 'combo_kernels': False, 'benchmark_combo_kernel': True, 'combo_kernel_foreach_dynamic_shapes': True}
+
+import torch._dynamo
+@torch._dynamo.disable
+def _call_8bit_base_layer(base_layer, x, *args, **kwargs):
+    return base_layer(x, *args, **kwargs)
+from torch import Tensor
+import torch
+import torch.nn as nn
+from torch.nn import functional as F
+from unsloth_zoo.temporary_patches.common import torch_compile
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+from peft.tuners.lora.bnb import (VARIANT_KWARG_KEYS, torch)
+
+
+torch_addmm = torch.addmm
+torch_add   = torch.add
+# @torch.compile(fullgraph = False, dynamic = True, options = torch_compile_options)
+def lora_forward(result, lora_A, lora_B, dropout, x, scaling):
+    # Use result.dtype (bfloat16 from base layer) since x may have been cast to float32
+    # by _cast_input_dtype when autocast is disabled
+    target_dtype = result.dtype
+    xA = dropout(x).to(target_dtype) @ lora_A.weight.to(target_dtype).t()
+    # output = result + scaling * xA @ lora_B.weight.t()
+    shape = result.shape
+    output = torch_addmm(
+        result.view(-1, shape[-1]),
+        xA.view(-1, xA.shape[-1]),
+        lora_B.weight.to(target_dtype).t(),
+        alpha = scaling,
+        beta = 1,
+    ).view(shape)
+
+    bias = lora_B.bias
+    if bias is not None:
+        output = torch_add(
+            output,
+            bias.to(target_dtype),
+            alpha = scaling,
+        )
+    return output
+pass
+
+def unsloth_forward(self, x: torch.Tensor, *args, **kwargs) -> torch.Tensor:
+    
+    adapter_names = kwargs.pop("adapter_names", None)
+    variant_kwargs = {k: kwargs.pop(k, None) for k in VARIANT_KWARG_KEYS}  # don't pass these to base_layer
+
+    if self.disable_adapters:
+        if self.merged:
+            self.unmerge()
+        result = _call_8bit_base_layer(self.base_layer, x, *args, **kwargs)
+    elif adapter_names is not None:
+        result = self._mixed_batch_forward(x, *args, adapter_names=adapter_names, **variant_kwargs, **kwargs)
+    elif self.merged:
+        result = _call_8bit_base_layer(self.base_layer, x, *args, **kwargs)
+    else:
+        result = _call_8bit_base_layer(self.base_layer, x, *args, **kwargs)
+        for active_adapter in self.active_adapters:
+            if active_adapter not in self.lora_A.keys():
+                continue
+            lora_A = self.lora_A[active_adapter]
+            lora_B = self.lora_B[active_adapter]
+            dropout = self.lora_dropout[active_adapter]
+            scaling = self.scaling[active_adapter]
+
+            requires_conversion = not torch.is_autocast_enabled()
+            if requires_conversion:
+                expected_dtype = result.dtype
+                x = self._cast_input_dtype(x, lora_A.weight.dtype)
+
+            if active_adapter not in self.lora_variant:  # vanilla LoRA
+                return lora_forward(result, lora_A, lora_B, dropout, x, scaling).to(result.dtype)
+                if requires_conversion:
+                    output = output.to(expected_dtype)
+                result = result + output
+            else:
+                result = self.lora_variant[active_adapter].forward(
+                    self,
+                    active_adapter=active_adapter,
+                    x=x,
+                    result=result,
+                    **variant_kwargs,
+                    **kwargs,
+                )
+                if requires_conversion:
+                    result = result.to(expected_dtype)
+
+    return result

unsloth_compiled_cache/Linear_peft_forward.py

@@ -0,0 +1,115 @@
+"""
+2026.6.7
+2026.6.9
+5.5.0
+1.7.0
+__UNSLOTH_VERSIONING__
+"""
+
+# Unsloth auto generated code
+# Copyright 2023-present Daniel Han-Chen, Michael Han-Chen & the Unsloth team. All rights reserved.
+#
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Lesser General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public License
+# along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+
+try:
+    from peft.tuners.lora.layer import VARIANT_KWARG_KEYS
+except ImportError:
+    VARIANT_KWARG_KEYS = ['alora_offsets']
+torch_compile_options = {'epilogue_fusion': True, 'max_autotune': False, 'shape_padding': True, 'trace.enabled': False, 'triton.cudagraphs': False, 'debug': False, 'dce': True, 'memory_planning': True, 'coordinate_descent_tuning': False, 'trace.graph_diagram': False, 'compile_threads': 4, 'group_fusion': True, 'disable_progress': True, 'verbose_progress': False, 'triton.multi_kernel': 0, 'triton.use_block_ptr': False, 'triton.enable_persistent_tma_matmul': True, 'triton.autotune_at_compile_time': False, 'triton.cooperative_reductions': False, 'cuda.compile_opt_level': '-O2', 'cuda.enable_cuda_lto': True, 'combo_kernels': False, 'benchmark_combo_kernel': True, 'combo_kernel_foreach_dynamic_shapes': True}
+from torch import Tensor
+import torch
+import torch.nn as nn
+from torch.nn import functional as F
+from unsloth_zoo.temporary_patches.common import torch_compile
+from typing import Any, List, Optional, Tuple, Union, Dict, Set, Callable
+from peft.tuners.lora.variants import (Any, torch)
+
+
+torch_addmm = torch.addmm
+torch_add   = torch.add
+# @torch.compile(fullgraph = False, dynamic = True, options = torch_compile_options)
+def lora_forward(result, lora_A, lora_B, dropout, x, scaling):
+    # Use result.dtype (bfloat16 from base layer) since x may have been cast to float32
+    # by _cast_input_dtype when autocast is disabled
+    target_dtype = result.dtype
+    xA = dropout(x).to(target_dtype) @ lora_A.weight.to(target_dtype).t()
+    # output = result + scaling * xA @ lora_B.weight.t()
+    shape = result.shape
+    output = torch_addmm(
+        result.view(-1, shape[-1]),
+        xA.view(-1, xA.shape[-1]),
+        lora_B.weight.to(target_dtype).t(),
+        alpha = scaling,
+        beta = 1,
+    ).view(shape)
+
+    bias = lora_B.bias
+    if bias is not None:
+        output = torch_add(
+            output,
+            bias.to(target_dtype),
+            alpha = scaling,
+        )
+    return output
+pass
+
+def unsloth_forward(self, x: torch.Tensor, *args: Any, **kwargs: Any) -> torch.Tensor:
+    
+    adapter_names = kwargs.pop("adapter_names", None)
+    variant_kwargs = {k: kwargs.pop(k, None) for k in VARIANT_KWARG_KEYS}  # don't pass these to base_layer
+
+    if self.disable_adapters:
+        if self.merged:
+            self.unmerge()
+        if not torch.is_autocast_enabled() and hasattr(self.base_layer, 'weight') and self.base_layer.weight is not None and not hasattr(self.base_layer.weight, 'quant_state') and x.dtype != self.base_layer.weight.dtype:
+            x = x.to(self.base_layer.weight.dtype)
+        result = self.base_layer(x, *args, **kwargs)
+    elif adapter_names is not None:
+        result = self._mixed_batch_forward(x, *args, adapter_names=adapter_names, **variant_kwargs, **kwargs)
+    elif self.merged:
+        if not torch.is_autocast_enabled() and hasattr(self.base_layer, 'weight') and self.base_layer.weight is not None and not hasattr(self.base_layer.weight, 'quant_state') and x.dtype != self.base_layer.weight.dtype:
+            x = x.to(self.base_layer.weight.dtype)
+        result = self.base_layer(x, *args, **kwargs)
+    else:
+        if not torch.is_autocast_enabled() and hasattr(self.base_layer, 'weight') and self.base_layer.weight is not None and not hasattr(self.base_layer.weight, 'quant_state') and x.dtype != self.base_layer.weight.dtype:
+            x = x.to(self.base_layer.weight.dtype)
+        result = self.base_layer(x, *args, **kwargs)
+        torch_result_dtype = result.dtype
+
+        lora_A_keys = self.lora_A.keys()
+        for active_adapter in self.active_adapters:
+            if active_adapter not in lora_A_keys:
+                continue
+
+            lora_A = self.lora_A[active_adapter]
+            lora_B = self.lora_B[active_adapter]
+            dropout = self.lora_dropout[active_adapter]
+            scaling = self.scaling[active_adapter]
+            if not torch.is_autocast_enabled(): result, x = result.to(lora_A.weight.dtype), x.to(lora_A.weight.dtype)
+            if active_adapter not in self.lora_variant:  # vanilla LoRA
+                return lora_forward(result, lora_A, lora_B, dropout, x, scaling).to(torch_result_dtype)
+            else:
+                result = self.lora_variant[active_adapter].forward(
+                    self,
+                    active_adapter=active_adapter,
+                    x=x,
+                    result=result,
+                    **variant_kwargs,
+                    **kwargs,
+                )
+
+        result = result.to(torch_result_dtype)
+
+    return result