Upload folder using huggingface_hub

README.md

@@ -1,25 +1,65 @@
 ---
-base_model: Qwen/Qwen3-4B-Instruct-2507
+base_model: unsloth/Qwen3-4B-Instruct-2507
+datasets:
+- u-10bei/structured_data_with_cot_dataset_512_v5
+language:
+- en
+license: apache-2.0
 library_name: peft
 pipeline_tag: text-generation
 tags:
+- qlora
 - lora
 - structured-output
 ---
 
-# structeval-lora (LoRA adapter)
+＜【課題】ここは自分で記入して下さい＞
 
-LoRA adapter for the StructEval-style structured output main competition.
+This repository provides a **LoRA adapter** fine-tuned from
+**unsloth/Qwen3-4B-Instruct-2507** using **QLoRA (4-bit, Unsloth)**.
 
-## Base model
-- `Qwen/Qwen3-4B-Instruct-2507`
+This repository contains **LoRA adapter weights only**.
+The base model must be loaded separately.
 
-## Training data (allowed list)
-- `daichira/structured-hard-sft-4k` (SFT)
+## Training Objective
 
-## Method
-- LoRA fine-tuning (completion-only: learns the final assistant output)
+This adapter is trained to improve **structured output accuracy**
+(JSON / YAML / XML / TOML / CSV).
 
-## Notes
-- This adapter was trained locally on macOS (MPS) as a quick end-to-end pipeline check.
-- For better scores, train longer on the full allowed datasets (and optionally DPO) and generate the final `inference.json` using the official inference workflow.
+Loss is applied only to the final assistant output,
+while intermediate reasoning (Chain-of-Thought) is masked.
+
+## Training Configuration
+
+- Base model: unsloth/Qwen3-4B-Instruct-2507
+- Method: QLoRA (4-bit)
+- Max sequence length: 256
+- Epochs: 1
+- Learning rate: 5e-05
+- LoRA: r=16, alpha=32
+
+## Usage
+
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from peft import PeftModel
+import torch
+
+base = "unsloth/Qwen3-4B-Instruct-2507"
+adapter = "your_id/your-repo"
+
+tokenizer = AutoTokenizer.from_pretrained(base)
+model = AutoModelForCausalLM.from_pretrained(
+    base,
+    torch_dtype=torch.float16,
+    device_map="auto",
+)
+model = PeftModel.from_pretrained(model, adapter)
+```
+
+## Sources & Terms (IMPORTANT)
+
+Training data: u-10bei/structured_data_with_cot_dataset_512_v5
+
+Dataset License: MIT License. This dataset is used and distributed under the terms of the MIT License.
+Compliance: Users must comply with the MIT license (including copyright notice) and the base model's original terms of use.

adapter_config.json

@@ -3,7 +3,7 @@
   "alpha_pattern": {},
   "arrow_config": null,
   "auto_mapping": null,
-  "base_model_name_or_path": "Qwen/Qwen3-4B-Instruct-2507",
+  "base_model_name_or_path": "unsloth/Qwen3-4B-Instruct-2507",
   "bias": "none",
   "corda_config": null,
   "ensure_weight_tying": false,
@@ -18,7 +18,7 @@
   "loftq_config": {},
   "lora_alpha": 32,
   "lora_bias": false,
-  "lora_dropout": 0.0,
+  "lora_dropout": 0.05,
   "megatron_config": null,
   "megatron_core": "megatron.core",
   "modules_to_save": null,
@@ -29,13 +29,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "q_proj",
-    "up_proj",
     "k_proj",
-    "down_proj",
-    "gate_proj",
     "o_proj",
-    "v_proj"
+    "q_proj",
+    "v_proj",
+    "gate_proj",
+    "up_proj",
+    "down_proj"
   ],
   "target_parameters": null,
   "task_type": "CAUSAL_LM",

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fd3b9a66dc6d6b977afa71f191fca485dae248e56d717b4e2b1d66a7addad6f1
+oid sha256:736ece3c4c8255bc354810158945800126dccff05fea5db4e20e5ce6d3f0542f
 size 132187888

tokenizer_config.json

@@ -21,8 +21,10 @@
     "<|video_pad|>"
   ],
   "is_local": false,
-  "model_max_length": 1010000,
-  "pad_token": "<|endoftext|>",
+  "model_max_length": 262144,
+  "model_specific_special_tokens": {},
+  "pad_token": "<|vision_pad|>",
+  "padding_side": "left",
   "split_special_tokens": false,
   "tokenizer_class": "Qwen2Tokenizer",
   "unk_token": null