Upload LoRA adapter (README written by author)

README.md

@@ -13,11 +13,7 @@ tags:
 - structured-output
 ---
 
-# test-v0-bf16
-
-This repository provides a fine-tuned model for the competition in matsuo-lab LLM lecture 2025.
-
-The model of this repo is trained with BF16 LoRA. Training configuration is the same as the distributed code.
+# test-v1
 
 This repository provides a **LoRA adapter** fine-tuned from
 **Qwen/Qwen3-4B-Instruct-2507** using **QLoRA (4-bit, Unsloth)**.
@@ -39,7 +35,7 @@ while intermediate reasoning (Chain-of-Thought) is masked.
 - Method: QLoRA (4-bit)
 - Max sequence length: 512
 - Epochs: 1
-- Learning rate: 1e-06
+- Learning rate: 5e-05
 - LoRA: r=64, alpha=128
 
 ## Usage
@@ -50,7 +46,7 @@ from peft import PeftModel
 import torch
 
 base = "Qwen/Qwen3-4B-Instruct-2507"
-adapter = "your_id/your-repo"
+adapter = "amu870/test-v1"
 
 tokenizer = AutoTokenizer.from_pretrained(base)
 model = AutoModelForCausalLM.from_pretrained(

adapter_config.json

@@ -33,13 +33,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "down_proj",
-    "gate_proj",
     "q_proj",
-    "k_proj",
-    "up_proj",
     "v_proj",
-    "o_proj"
+    "down_proj",
+    "k_proj",
+    "o_proj",
+    "gate_proj",
+    "up_proj"
   ],
   "target_parameters": null,
   "task_type": "CAUSAL_LM",

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3ca6921351d4d50235aa25570855384cf3090ee27427f8a545bbe99ace40cd89
+oid sha256:845894e012cd035e2f326f438c238d734fe713d24619cc99f522c27d30d1b08b
 size 528550256