Instructions to use tsavage68/500STEPS_1e6rate_01beta_DPO_Meditron7B_zeroshot with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use tsavage68/500STEPS_1e6rate_01beta_DPO_Meditron7B_zeroshot with Transformers:

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

pipe = pipeline("text-generation", model="tsavage68/500STEPS_1e6rate_01beta_DPO_Meditron7B_zeroshot")

# Load model directly
from transformers import AutoTokenizer, AutoModelForMultimodalLM

tokenizer = AutoTokenizer.from_pretrained("tsavage68/500STEPS_1e6rate_01beta_DPO_Meditron7B_zeroshot")
model = AutoModelForMultimodalLM.from_pretrained("tsavage68/500STEPS_1e6rate_01beta_DPO_Meditron7B_zeroshot")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use tsavage68/500STEPS_1e6rate_01beta_DPO_Meditron7B_zeroshot with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "tsavage68/500STEPS_1e6rate_01beta_DPO_Meditron7B_zeroshot"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "tsavage68/500STEPS_1e6rate_01beta_DPO_Meditron7B_zeroshot",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/tsavage68/500STEPS_1e6rate_01beta_DPO_Meditron7B_zeroshot

SGLang

How to use tsavage68/500STEPS_1e6rate_01beta_DPO_Meditron7B_zeroshot 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 "tsavage68/500STEPS_1e6rate_01beta_DPO_Meditron7B_zeroshot" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "tsavage68/500STEPS_1e6rate_01beta_DPO_Meditron7B_zeroshot",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "tsavage68/500STEPS_1e6rate_01beta_DPO_Meditron7B_zeroshot" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "tsavage68/500STEPS_1e6rate_01beta_DPO_Meditron7B_zeroshot",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use tsavage68/500STEPS_1e6rate_01beta_DPO_Meditron7B_zeroshot with Docker Model Runner:
```
docker model run hf.co/tsavage68/500STEPS_1e6rate_01beta_DPO_Meditron7B_zeroshot
```

500STEPS_1e6rate_01beta_DPO_Meditron7B

This model is a fine-tuned version of epfl-llm/meditron-7b on an unknown dataset. It achieves the following results on the evaluation set:

Loss: 0.6302
Rewards/chosen: 0.0115
Rewards/rejected: -0.1672
Rewards/accuracies: 0.5868
Rewards/margins: 0.1788
Logps/rejected: -29.4661
Logps/chosen: -26.3659
Logits/rejected: -0.7645
Logits/chosen: -0.7643

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-06
train_batch_size: 4
eval_batch_size: 1
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 8
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 100
training_steps: 500

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6902	0.1	50	0.6903	0.0090	0.0031	0.5121	0.0058	-27.7623	-26.3918	-0.6125	-0.6124
0.6766	0.2	100	0.6792	-0.1559	-0.1907	0.5099	0.0349	-29.7009	-28.0399	-0.6382	-0.6380
0.6667	0.29	150	0.6567	-0.0224	-0.1102	0.5714	0.0879	-28.8959	-26.7051	-0.6559	-0.6557
0.6656	0.39	200	0.6495	-0.0303	-0.1387	0.5802	0.1084	-29.1808	-26.7847	-0.7108	-0.7106
0.5939	0.49	250	0.6388	-0.0202	-0.1629	0.5890	0.1426	-29.4223	-26.6837	-0.7329	-0.7327
0.6328	0.59	300	0.6349	-0.0421	-0.2022	0.5758	0.1601	-29.8158	-26.9024	-0.7492	-0.7490
0.6231	0.68	350	0.6313	-0.0004	-0.1725	0.5758	0.1721	-29.5189	-26.4852	-0.7571	-0.7569
0.6419	0.78	400	0.6303	0.0123	-0.1660	0.5868	0.1783	-29.4536	-26.3585	-0.7639	-0.7637
0.6045	0.88	450	0.6304	0.0120	-0.1662	0.5846	0.1783	-29.4560	-26.3611	-0.7645	-0.7643
0.5984	0.98	500	0.6302	0.0115	-0.1672	0.5868	0.1788	-29.4661	-26.3659	-0.7645	-0.7643

Framework versions

Transformers 4.37.2
Pytorch 2.0.0+cu117
Datasets 2.17.0
Tokenizers 0.15.1

Downloads last month: 7

Safetensors

Model size

7B params

Tensor type

F16

Model tree for tsavage68/500STEPS_1e6rate_01beta_DPO_Meditron7B_zeroshot

Base model

meta-llama/Llama-2-7b

Finetuned

epfl-llm/meditron-7b

Finetuned

(41)

this model