Instructions to use tsavage68/UTI2_M2_1000steps_1e7rate_CSFTDPO with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use tsavage68/UTI2_M2_1000steps_1e7rate_CSFTDPO with Transformers:

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

pipe = pipeline("text-generation", model="tsavage68/UTI2_M2_1000steps_1e7rate_CSFTDPO")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("tsavage68/UTI2_M2_1000steps_1e7rate_CSFTDPO")
model = AutoModelForCausalLM.from_pretrained("tsavage68/UTI2_M2_1000steps_1e7rate_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/UTI2_M2_1000steps_1e7rate_CSFTDPO with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "tsavage68/UTI2_M2_1000steps_1e7rate_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/UTI2_M2_1000steps_1e7rate_CSFTDPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/tsavage68/UTI2_M2_1000steps_1e7rate_CSFTDPO

SGLang

How to use tsavage68/UTI2_M2_1000steps_1e7rate_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/UTI2_M2_1000steps_1e7rate_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/UTI2_M2_1000steps_1e7rate_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/UTI2_M2_1000steps_1e7rate_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/UTI2_M2_1000steps_1e7rate_CSFTDPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use tsavage68/UTI2_M2_1000steps_1e7rate_CSFTDPO with Docker Model Runner:
```
docker model run hf.co/tsavage68/UTI2_M2_1000steps_1e7rate_CSFTDPO
```

UTI2_M2_1000steps_1e7rate_CSFTDPO

This model is a fine-tuned version of tsavage68/UTI_M2_1000steps_1e7rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:

Loss: 0.5546
Rewards/chosen: 0.0422
Rewards/rejected: -0.2698
Rewards/accuracies: 0.8600
Rewards/margins: 0.3120
Logps/rejected: -39.8957
Logps/chosen: -19.8371
Logits/rejected: -2.6809
Logits/chosen: -2.6783

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
0.6931	0.3333	25	0.6921	0.0031	0.0001	0.1900	0.0030	-39.3560	-19.9153	-2.6832	-2.6806
0.6879	0.6667	50	0.6797	0.0157	-0.0141	0.5300	0.0298	-39.3843	-19.8902	-2.6825	-2.6799
0.7047	1.0	75	0.6833	0.0064	-0.0173	0.5300	0.0237	-39.3907	-19.9087	-2.6821	-2.6796
0.6925	1.3333	100	0.6719	0.0150	-0.0314	0.6200	0.0464	-39.4189	-19.8915	-2.6833	-2.6807
0.6674	1.6667	125	0.6638	0.0030	-0.0600	0.6900	0.0630	-39.4762	-19.9155	-2.6817	-2.6791
0.6591	2.0	150	0.6356	0.0148	-0.1082	0.8100	0.1230	-39.5726	-19.8920	-2.6816	-2.6790
0.637	2.3333	175	0.6319	0.0113	-0.1196	0.8200	0.1309	-39.5954	-19.8989	-2.6812	-2.6786
0.6179	2.6667	200	0.6054	0.0342	-0.1567	0.8300	0.1909	-39.6696	-19.8532	-2.6821	-2.6795
0.6173	3.0	225	0.6032	0.0393	-0.1577	0.8200	0.1970	-39.6716	-19.8429	-2.6816	-2.6790
0.5873	3.3333	250	0.5858	0.0189	-0.2169	0.8400	0.2358	-39.7899	-19.8837	-2.6812	-2.6786
0.5795	3.6667	275	0.5877	0.0141	-0.2185	0.8000	0.2326	-39.7932	-19.8934	-2.6813	-2.6787
0.6008	4.0	300	0.5756	0.0356	-0.2244	0.8400	0.2600	-39.8049	-19.8503	-2.6803	-2.6777
0.57	4.3333	325	0.5764	0.0262	-0.2323	0.8400	0.2585	-39.8208	-19.8692	-2.6807	-2.6781
0.5584	4.6667	350	0.5605	0.0242	-0.2723	0.8600	0.2964	-39.9007	-19.8732	-2.6802	-2.6776
0.572	5.0	375	0.5604	0.0279	-0.2703	0.8700	0.2982	-39.8968	-19.8658	-2.6804	-2.6778
0.5811	5.3333	400	0.5617	0.0342	-0.2607	0.8500	0.2949	-39.8776	-19.8531	-2.6798	-2.6772
0.5751	5.6667	425	0.5648	0.0392	-0.2472	0.8600	0.2865	-39.8506	-19.8431	-2.6809	-2.6783
0.561	6.0	450	0.5624	0.0124	-0.2803	0.8500	0.2927	-39.9167	-19.8967	-2.6806	-2.6781
0.545	6.3333	475	0.5525	0.0448	-0.2732	0.8700	0.3180	-39.9025	-19.8319	-2.6815	-2.6789
0.6125	6.6667	500	0.5589	0.0463	-0.2561	0.8700	0.3023	-39.8683	-19.8290	-2.6811	-2.6785
0.5398	7.0	525	0.5612	0.0214	-0.2753	0.8400	0.2966	-39.9067	-19.8788	-2.6805	-2.6779
0.543	7.3333	550	0.5643	0.0400	-0.2494	0.8500	0.2894	-39.8549	-19.8415	-2.6806	-2.6781
0.5541	7.6667	575	0.5616	0.0247	-0.2721	0.8500	0.2968	-39.9002	-19.8720	-2.6813	-2.6788
0.5576	8.0	600	0.5650	0.0122	-0.2764	0.8500	0.2886	-39.9089	-19.8971	-2.6812	-2.6786
0.5543	8.3333	625	0.5605	0.0330	-0.2649	0.8600	0.2980	-39.8860	-19.8555	-2.6809	-2.6783
0.5405	8.6667	650	0.5648	0.0146	-0.2732	0.8500	0.2878	-39.9025	-19.8924	-2.6810	-2.6784
0.5535	9.0	675	0.5536	0.0354	-0.2789	0.8500	0.3143	-39.9140	-19.8507	-2.6798	-2.6772
0.5292	9.3333	700	0.5534	0.0444	-0.2708	0.8600	0.3152	-39.8978	-19.8328	-2.6808	-2.6782
0.5718	9.6667	725	0.5556	0.0429	-0.2668	0.8400	0.3097	-39.8898	-19.8358	-2.6813	-2.6787
0.585	10.0	750	0.5512	0.0392	-0.2799	0.8800	0.3191	-39.9159	-19.8431	-2.6809	-2.6783
0.5609	10.3333	775	0.5540	0.0352	-0.2800	0.8600	0.3152	-39.9161	-19.8511	-2.6808	-2.6782
0.5572	10.6667	800	0.5500	0.0424	-0.2816	0.8700	0.3240	-39.9193	-19.8367	-2.6809	-2.6783
0.5514	11.0	825	0.5541	0.0433	-0.2698	0.8700	0.3131	-39.8958	-19.8350	-2.6809	-2.6783
0.5467	11.3333	850	0.5546	0.0422	-0.2698	0.8600	0.3120	-39.8957	-19.8371	-2.6809	-2.6783
0.5803	11.6667	875	0.5546	0.0422	-0.2698	0.8600	0.3120	-39.8957	-19.8371	-2.6809	-2.6783
0.5514	12.0	900	0.5546	0.0422	-0.2698	0.8600	0.3120	-39.8957	-19.8371	-2.6809	-2.6783
0.5579	12.3333	925	0.5546	0.0422	-0.2698	0.8600	0.3120	-39.8957	-19.8371	-2.6809	-2.6783
0.5599	12.6667	950	0.5546	0.0422	-0.2698	0.8600	0.3120	-39.8957	-19.8371	-2.6809	-2.6783
0.5609	13.0	975	0.5546	0.0422	-0.2698	0.8600	0.3120	-39.8957	-19.8371	-2.6809	-2.6783
0.552	13.3333	1000	0.5546	0.0422	-0.2698	0.8600	0.3120	-39.8957	-19.8371	-2.6809	-2.6783

Framework versions

Transformers 4.41.2
Pytorch 2.0.0+cu117
Datasets 2.19.2
Tokenizers 0.19.1

Downloads last month: 6

Safetensors

Model size

7B params

Tensor type

F16

Model tree for tsavage68/UTI2_M2_1000steps_1e7rate_CSFTDPO

Base model

mistralai/Mistral-7B-Instruct-v0.2

Finetuned

tsavage68/UTI_M2_1000steps_1e7rate_SFT

Finetuned

(32)

this model