Instructions to use Physics-Synthetic-Data/Qwen2.5-coder-7B-Dec20-SFT with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Physics-Synthetic-Data/Qwen2.5-coder-7B-Dec20-SFT with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="Physics-Synthetic-Data/Qwen2.5-coder-7B-Dec20-SFT")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Physics-Synthetic-Data/Qwen2.5-coder-7B-Dec20-SFT")
model = AutoModelForCausalLM.from_pretrained("Physics-Synthetic-Data/Qwen2.5-coder-7B-Dec20-SFT")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use Physics-Synthetic-Data/Qwen2.5-coder-7B-Dec20-SFT with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Physics-Synthetic-Data/Qwen2.5-coder-7B-Dec20-SFT"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Physics-Synthetic-Data/Qwen2.5-coder-7B-Dec20-SFT",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Physics-Synthetic-Data/Qwen2.5-coder-7B-Dec20-SFT

SGLang

How to use Physics-Synthetic-Data/Qwen2.5-coder-7B-Dec20-SFT with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Physics-Synthetic-Data/Qwen2.5-coder-7B-Dec20-SFT" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Physics-Synthetic-Data/Qwen2.5-coder-7B-Dec20-SFT",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Physics-Synthetic-Data/Qwen2.5-coder-7B-Dec20-SFT" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Physics-Synthetic-Data/Qwen2.5-coder-7B-Dec20-SFT",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Physics-Synthetic-Data/Qwen2.5-coder-7B-Dec20-SFT with Docker Model Runner:
```
docker model run hf.co/Physics-Synthetic-Data/Qwen2.5-coder-7B-Dec20-SFT
```

See axolotl config

axolotl version: 0.10.0

# ---------------- 核心模型路径 ----------------
base_model: /nfs/data/johnsonk/mllm_models/RAFT/Oct2/lr1e-5_epoch4_warm0.03_training_6k_GPU_0_1_2_3_20251001_172635/trained_model   
# output_dir: /work/nvme/bckr/wli18/RAFT/exp_1
dataset_prepared_path: last_run_prepared
# --------------- 数据集 ----------------------
datasets:
  # 训练集
  - path: /home/johnsonk/RAFT/1exp/exp_1.0.2_questionize_high_school_physics_861_img/result/merged_physics_tikz_dataset.jsonl
    type: chat_template
    split: train
    field_messages: messages
    message_property_mappings:
      role: role
      content: content
val_set_size: 0.05    

# --------------- 训练超参 ---------------------

sequence_len: 2048
micro_batch_size: 1 
gradient_accumulation_steps: 8
num_epochs: 3 
learning_rate: 1e-5
optimizer: adamw_torch
lr_scheduler: cosine
warmup_ratio: 0.03
weight_decay: 0.01



# max_steps: 100  # <--- 新增此行，强制运行100步
# --------------- 精度 / 显存 ------------------
bf16: auto
tf32: true
gradient_checkpointing: true
# gradient_checkpointing_kwargs:
#   use_reentrant: false

# --------------- DeepSpeed --------------------
deepspeed: /home/johnsonk/RAFT/configs/deepspeed_stage3.json

# ----------- 关闭所有 LoRA / 量化 -------------
# (配置已为全参数训练)
# load_in_4bit: true
# adapter: qlora
# lora_r: 16
# lora_alpha: 32
# lora_target_modules:
#   - q_proj
#   - k_proj
#   - v_proj
#   - o_proj
#   - down_proj
#   - up_proj
# lora_mlp_kernel: true
# lora_qkv_kernel: true
# lora_o_kernel: true

# -------------- 其它杂项 ----------------------
strict: false
chat_template: qwen_25
save_steps: 100000  # 每50个step保存一次

# [关键] 设置最多只保留2个最新的checkpoint，自动删除旧的
save_total_limit: 0

evals_per_epoch: 1
# saves_per_epoch: 100
logging_steps: 1
flash_attention: true


### WandB ###
report_to: wandb
wandb_project: test_axolotl
wandb_entity: johnson0213-ucla
wandb_name: Qwen2.5-Coder-7B-TikZ-SFT-860-v1-3epoch-grad_acc8-lr1e-5

nfs/data/johnsonk/mllm_models/RAFT/Dec20/lr5e-6_epoch16_warm0.03_training_860_GPU_0_1_2_3_20251220_021048/trained_model

This model was trained from scratch on the /home/johnsonk/RAFT/1exp/exp_1.0.2_questionize_high_school_physics_861_img/result/merged_physics_tikz_dataset.jsonl dataset. It achieves the following results on the evaluation set:

Loss: 0.3683

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 1e-05
train_batch_size: 1
eval_batch_size: 1
seed: 42
distributed_type: multi-GPU
num_devices: 4
gradient_accumulation_steps: 8
total_train_batch_size: 32
total_eval_batch_size: 4
optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 2
training_steps: 77

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0	0	1.6350
0.2283	1.0	26	0.3596
0.1082	2.0	52	0.3683

Framework versions

Transformers 4.52.3
Pytorch 2.5.1
Datasets 3.6.0
Tokenizers 0.21.2

Downloads last month: 2

Safetensors

Model size

8B params

Tensor type

BF16