Instructions to use tsavage68/MedQA_L3_1000steps_1e8rate_01beta_CSFTDPO with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use tsavage68/MedQA_L3_1000steps_1e8rate_01beta_CSFTDPO with Transformers:

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

pipe = pipeline("text-generation", model="tsavage68/MedQA_L3_1000steps_1e8rate_01beta_CSFTDPO")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("tsavage68/MedQA_L3_1000steps_1e8rate_01beta_CSFTDPO")
model = AutoModelForCausalLM.from_pretrained("tsavage68/MedQA_L3_1000steps_1e8rate_01beta_CSFTDPO", device_map="auto")
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 Settings

vLLM

How to use tsavage68/MedQA_L3_1000steps_1e8rate_01beta_CSFTDPO with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "tsavage68/MedQA_L3_1000steps_1e8rate_01beta_CSFTDPO"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "tsavage68/MedQA_L3_1000steps_1e8rate_01beta_CSFTDPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/tsavage68/MedQA_L3_1000steps_1e8rate_01beta_CSFTDPO

SGLang

How to use tsavage68/MedQA_L3_1000steps_1e8rate_01beta_CSFTDPO 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 "tsavage68/MedQA_L3_1000steps_1e8rate_01beta_CSFTDPO" \
    --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": "tsavage68/MedQA_L3_1000steps_1e8rate_01beta_CSFTDPO",
		"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 "tsavage68/MedQA_L3_1000steps_1e8rate_01beta_CSFTDPO" \
        --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": "tsavage68/MedQA_L3_1000steps_1e8rate_01beta_CSFTDPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use tsavage68/MedQA_L3_1000steps_1e8rate_01beta_CSFTDPO with Docker Model Runner:
```
docker model run hf.co/tsavage68/MedQA_L3_1000steps_1e8rate_01beta_CSFTDPO
```

MedQA_L3_1000steps_1e8rate_01beta_CSFTDPO

This model is a fine-tuned version of tsavage68/MedQA_L3_1000steps_1e6rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:

Loss: 0.6933
Rewards/chosen: -0.0003
Rewards/rejected: -0.0001
Rewards/accuracies: 0.4923
Rewards/margins: -0.0002
Logps/rejected: -33.8557
Logps/chosen: -31.3318
Logits/rejected: -0.7327
Logits/chosen: -0.7320

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-08
train_batch_size: 2
eval_batch_size: 1
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 4
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 100
training_steps: 1000

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6929	0.0489	50	0.6932	0.0003	0.0004	0.4967	-0.0001	-33.8504	-31.3255	-0.7322	-0.7315
0.6934	0.0977	100	0.6934	0.0004	0.0009	0.4703	-0.0005	-33.8457	-31.3248	-0.7326	-0.7320
0.6924	0.1466	150	0.6931	0.0051	0.0049	0.5165	0.0002	-33.8057	-31.2774	-0.7323	-0.7316
0.6943	0.1954	200	0.6928	0.0019	0.0012	0.5099	0.0008	-33.8433	-31.3093	-0.7327	-0.7320
0.6931	0.2443	250	0.6930	0.0022	0.0018	0.5055	0.0004	-33.8372	-31.3066	-0.7324	-0.7317
0.6948	0.2931	300	0.6928	0.0049	0.0041	0.5275	0.0008	-33.8138	-31.2796	-0.7324	-0.7318
0.6952	0.3420	350	0.6932	0.0015	0.0015	0.4571	0.0000	-33.8399	-31.3133	-0.7327	-0.7321
0.694	0.3908	400	0.6932	0.0018	0.0019	0.4791	-0.0002	-33.8358	-31.3110	-0.7326	-0.7319
0.6941	0.4397	450	0.6932	-0.0010	-0.0009	0.5033	-0.0001	-33.8636	-31.3385	-0.7322	-0.7315
0.6919	0.4885	500	0.6933	0.0032	0.0034	0.4945	-0.0002	-33.8206	-31.2967	-0.7322	-0.7316
0.6955	0.5374	550	0.6934	0.0013	0.0018	0.4989	-0.0005	-33.8370	-31.3153	-0.7324	-0.7317
0.6915	0.5862	600	0.6931	0.0004	0.0003	0.5253	0.0001	-33.8517	-31.3242	-0.7327	-0.7320
0.6911	0.6351	650	0.6935	0.0005	0.0011	0.4703	-0.0006	-33.8438	-31.3237	-0.7325	-0.7318
0.6921	0.6839	700	0.6930	-0.0015	-0.0019	0.5165	0.0004	-33.8742	-31.3438	-0.7324	-0.7318
0.6926	0.7328	750	0.6931	0.0012	0.0011	0.5187	0.0001	-33.8440	-31.3166	-0.7328	-0.7321
0.6927	0.7816	800	0.6930	0.0018	0.0014	0.5143	0.0004	-33.8407	-31.3102	-0.7325	-0.7318
0.6949	0.8305	850	0.6933	-0.0003	-0.0001	0.4901	-0.0003	-33.8555	-31.3320	-0.7327	-0.7320
0.6942	0.8793	900	0.6933	-0.0003	-0.0001	0.4923	-0.0002	-33.8557	-31.3318	-0.7327	-0.7320
0.691	0.9282	950	0.6933	-0.0003	-0.0001	0.4923	-0.0002	-33.8557	-31.3318	-0.7327	-0.7320
0.6926	0.9770	1000	0.6933	-0.0003	-0.0001	0.4923	-0.0002	-33.8557	-31.3318	-0.7327	-0.7320

Framework versions

Transformers 4.41.1
Pytorch 2.0.0+cu117
Datasets 2.19.1
Tokenizers 0.19.1

Downloads last month: 11

Safetensors

Model size

8B params

Tensor type

F16

Model tree for tsavage68/MedQA_L3_1000steps_1e8rate_01beta_CSFTDPO

Base model

meta-llama/Meta-Llama-3-8B-Instruct

Finetuned

tsavage68/MedQA_L3_1000steps_1e6rate_SFT

Finetuned

(20)

this model