Instructions to use tsavage68/UTI2_M2_1000steps_1e7rate_05beta_CSFTDPO with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use tsavage68/UTI2_M2_1000steps_1e7rate_05beta_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_05beta_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_05beta_CSFTDPO")
model = AutoModelForCausalLM.from_pretrained("tsavage68/UTI2_M2_1000steps_1e7rate_05beta_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_05beta_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_05beta_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_05beta_CSFTDPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/tsavage68/UTI2_M2_1000steps_1e7rate_05beta_CSFTDPO

SGLang

How to use tsavage68/UTI2_M2_1000steps_1e7rate_05beta_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_05beta_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_05beta_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_05beta_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_05beta_CSFTDPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use tsavage68/UTI2_M2_1000steps_1e7rate_05beta_CSFTDPO with Docker Model Runner:
```
docker model run hf.co/tsavage68/UTI2_M2_1000steps_1e7rate_05beta_CSFTDPO
```

UTI2_M2_1000steps_1e7rate_05beta_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: 0.5476
Rewards/chosen: 0.0699
Rewards/rejected: -3.0830
Rewards/accuracies: 0.2100
Rewards/margins: 3.1530
Logps/rejected: -15.5400
Logps/chosen: -4.4026
Logits/rejected: -2.6403
Logits/chosen: -2.6398

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-07
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
0.7009	0.3333	25	0.6667	0.0016	-0.0631	0.1900	0.0647	-9.5002	-4.5393	-2.7063	-2.7055
0.5794	0.6667	50	0.5690	0.0605	-0.5674	0.2100	0.6279	-10.5088	-4.4215	-2.6950	-2.6942
0.5772	1.0	75	0.5638	-0.0374	-1.8778	0.2000	1.8404	-13.1295	-4.6173	-2.6653	-2.6647
0.5715	1.3333	100	0.5485	0.0321	-2.2707	0.2100	2.3028	-13.9154	-4.4783	-2.6560	-2.6555
0.5545	1.6667	125	0.5476	0.1013	-2.5349	0.2100	2.6363	-14.4438	-4.3398	-2.6499	-2.6494
0.5545	2.0	150	0.5476	0.0902	-2.9376	0.2100	3.0278	-15.2492	-4.3621	-2.6442	-2.6437
0.5545	2.3333	175	0.5476	0.0846	-2.9244	0.2100	3.0090	-15.2229	-4.3733	-2.6424	-2.6419
0.4852	2.6667	200	0.5476	0.0848	-2.9648	0.2100	3.0495	-15.3035	-4.3729	-2.6423	-2.6417
0.6412	3.0	225	0.5476	0.0853	-2.9694	0.2100	3.0547	-15.3127	-4.3718	-2.6421	-2.6415
0.5545	3.3333	250	0.5476	0.0892	-2.9671	0.2100	3.0563	-15.3081	-4.3640	-2.6429	-2.6424
0.5372	3.6667	275	0.5476	0.0803	-2.9507	0.2100	3.0310	-15.2754	-4.3819	-2.6416	-2.6410
0.5892	4.0	300	0.5476	0.0791	-3.0080	0.2100	3.0871	-15.3899	-4.3842	-2.6421	-2.6415
0.4679	4.3333	325	0.5476	0.0770	-3.0043	0.2100	3.0814	-15.3826	-4.3884	-2.6420	-2.6415
0.5718	4.6667	350	0.5476	0.0767	-3.0040	0.2100	3.0808	-15.3820	-4.3890	-2.6414	-2.6409
0.5199	5.0	375	0.5476	0.0830	-3.0444	0.2100	3.1274	-15.4628	-4.3765	-2.6415	-2.6410
0.5025	5.3333	400	0.5476	0.0784	-3.0520	0.2100	3.1304	-15.4779	-4.3857	-2.6406	-2.6401
0.5199	5.6667	425	0.5476	0.0772	-3.0417	0.2100	3.1189	-15.4575	-4.3882	-2.6418	-2.6412
0.5025	6.0	450	0.5476	0.0775	-3.0690	0.2100	3.1465	-15.5119	-4.3875	-2.6403	-2.6398
0.5718	6.3333	475	0.5476	0.0722	-3.0608	0.2100	3.1330	-15.4956	-4.3980	-2.6403	-2.6398
0.5718	6.6667	500	0.5476	0.0733	-3.0661	0.2100	3.1394	-15.5061	-4.3958	-2.6403	-2.6397
0.5025	7.0	525	0.5476	0.0687	-3.0692	0.2100	3.1379	-15.5123	-4.4051	-2.6407	-2.6402
0.5199	7.3333	550	0.5476	0.0691	-3.0762	0.2100	3.1454	-15.5265	-4.4042	-2.6401	-2.6396
0.5372	7.6667	575	0.5476	0.0728	-3.0945	0.2100	3.1672	-15.5629	-4.3970	-2.6414	-2.6409
0.5718	8.0	600	0.5476	0.0736	-3.0806	0.2100	3.1541	-15.5351	-4.3953	-2.6405	-2.6400
0.5372	8.3333	625	0.5476	0.0806	-3.0954	0.2100	3.1759	-15.5647	-4.3813	-2.6410	-2.6405
0.4332	8.6667	650	0.5476	0.0762	-3.0922	0.2100	3.1684	-15.5583	-4.3900	-2.6412	-2.6407
0.5372	9.0	675	0.5476	0.0738	-3.0924	0.2100	3.1662	-15.5587	-4.3948	-2.6408	-2.6403
0.5025	9.3333	700	0.5476	0.0702	-3.0892	0.2100	3.1594	-15.5524	-4.4020	-2.6405	-2.6400
0.5025	9.6667	725	0.5476	0.0641	-3.0956	0.2100	3.1597	-15.5651	-4.4142	-2.6410	-2.6405
0.5892	10.0	750	0.5476	0.0696	-3.0933	0.2100	3.1630	-15.5606	-4.4032	-2.6403	-2.6398
0.5199	10.3333	775	0.5476	0.0764	-3.0810	0.2100	3.1574	-15.5361	-4.3897	-2.6404	-2.6399
0.5199	10.6667	800	0.5476	0.0750	-3.0945	0.2100	3.1695	-15.5629	-4.3925	-2.6399	-2.6394
0.5372	11.0	825	0.5477	0.0727	-3.0777	0.2100	3.1504	-15.5293	-4.3970	-2.6405	-2.6399
0.5199	11.3333	850	0.5477	0.0760	-3.0775	0.2100	3.1534	-15.5289	-4.3905	-2.6402	-2.6397
0.6065	11.6667	875	0.5476	0.0737	-3.0877	0.2100	3.1615	-15.5495	-4.3950	-2.6404	-2.6398
0.5718	12.0	900	0.5476	0.0713	-3.0915	0.2100	3.1628	-15.5570	-4.3999	-2.6403	-2.6398
0.4159	12.3333	925	0.5476	0.0687	-3.0820	0.2100	3.1507	-15.5379	-4.4051	-2.6403	-2.6398
0.6238	12.6667	950	0.5476	0.0699	-3.0830	0.2100	3.1530	-15.5400	-4.4026	-2.6403	-2.6398
0.6065	13.0	975	0.5476	0.0699	-3.0830	0.2100	3.1530	-15.5400	-4.4026	-2.6403	-2.6398
0.5025	13.3333	1000	0.5476	0.0699	-3.0830	0.2100	3.1530	-15.5400	-4.4026	-2.6403	-2.6398

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_05beta_CSFTDPO

Base model

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

Finetuned

tsavage68/UTI_M2_1000steps_1e7rate_SFT

Finetuned

(32)

this model