Instructions to use tsavage68/UTI2_M2_1000steps_1e6rate_01beta_CSFTDPO with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use tsavage68/UTI2_M2_1000steps_1e6rate_01beta_CSFTDPO with Transformers:

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

pipe = pipeline("text-generation", model="tsavage68/UTI2_M2_1000steps_1e6rate_01beta_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_1e6rate_01beta_CSFTDPO")
model = AutoModelForCausalLM.from_pretrained("tsavage68/UTI2_M2_1000steps_1e6rate_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/UTI2_M2_1000steps_1e6rate_01beta_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_1e6rate_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/UTI2_M2_1000steps_1e6rate_01beta_CSFTDPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/tsavage68/UTI2_M2_1000steps_1e6rate_01beta_CSFTDPO

SGLang

How to use tsavage68/UTI2_M2_1000steps_1e6rate_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/UTI2_M2_1000steps_1e6rate_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/UTI2_M2_1000steps_1e6rate_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/UTI2_M2_1000steps_1e6rate_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/UTI2_M2_1000steps_1e6rate_01beta_CSFTDPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use tsavage68/UTI2_M2_1000steps_1e6rate_01beta_CSFTDPO with Docker Model Runner:
```
docker model run hf.co/tsavage68/UTI2_M2_1000steps_1e6rate_01beta_CSFTDPO
```

UTI2_M2_1000steps_1e6rate_01beta_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: 0.2525
Rewards/chosen: -4.6992
Rewards/rejected: -18.9442
Rewards/accuracies: 0.8800
Rewards/margins: 14.2450
Logps/rejected: -228.7986
Logps/chosen: -66.9138
Logits/rejected: -1.9681
Logits/chosen: -2.0652

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: 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
0.1981	0.3333	25	0.1707	0.4309	-7.1069	0.8800	7.5378	-110.4247	-15.6125	-2.4808	-2.4822
0.5206	0.6667	50	0.0694	0.4032	-9.1284	0.9000	9.5316	-130.6399	-15.8895	-2.4773	-2.4842
0.0347	1.0	75	0.3666	-3.1187	-8.7205	0.7800	5.6018	-126.5613	-51.1087	-2.4899	-2.4995
0.0788	1.3333	100	0.4184	0.4123	-15.2113	0.8800	15.6236	-191.4688	-15.7981	-2.4522	-2.4834
0.052	1.6667	125	0.4180	0.4111	-15.2137	0.8800	15.6248	-191.4934	-15.8105	-2.4523	-2.4835
0.1213	2.0	150	0.3560	-3.5267	-16.7763	0.8800	13.2496	-207.1189	-55.1884	-1.9818	-2.0298
0.2125	2.3333	175	0.1987	-5.3431	-19.4504	0.8900	14.1074	-233.8604	-73.3521	-2.0857	-2.1454
0.5195	2.6667	200	0.2487	-5.1181	-17.0835	0.8700	11.9655	-210.1916	-71.1022	-1.9512	-2.0125
0.0347	3.0	225	0.5257	-4.6932	-16.4184	0.8700	11.7251	-203.5397	-66.8538	-1.9297	-1.9821
0.0347	3.3333	250	0.5252	-4.6368	-16.5979	0.8700	11.9611	-205.3350	-66.2896	-1.9309	-1.9836
0.0693	3.6667	275	0.5277	-4.6207	-16.6952	0.8700	12.0744	-206.3079	-66.1288	-1.9307	-1.9833
0.3528	4.0	300	0.2783	-5.2353	-17.9334	0.8800	12.6980	-218.6897	-72.2747	-2.1007	-2.2018
0.0866	4.3333	325	0.5393	-5.1111	-18.7048	0.8700	13.5937	-226.4044	-71.0329	-2.0012	-2.0945
0.0347	4.6667	350	0.5419	-5.1092	-18.7024	0.8700	13.5933	-226.3804	-71.0133	-2.0016	-2.0948
0.0693	5.0	375	0.2439	-4.7227	-18.8694	0.8900	14.1466	-228.0500	-67.1490	-1.9662	-2.0633
0.0693	5.3333	400	0.2601	-4.7346	-18.8758	0.8800	14.1412	-228.1138	-67.2672	-1.9665	-2.0635
0.052	5.6667	425	0.2475	-4.7163	-18.8815	0.8800	14.1653	-228.1716	-67.0841	-1.9664	-2.0634
0.0866	6.0	450	0.2467	-4.7194	-18.8925	0.8800	14.1731	-228.2814	-67.1153	-1.9668	-2.0638
0.0173	6.3333	475	0.2504	-4.7284	-18.8991	0.8800	14.1708	-228.3476	-67.2055	-1.9666	-2.0636
0.1386	6.6667	500	0.2482	-4.7124	-18.9127	0.8800	14.2002	-228.4828	-67.0460	-1.9668	-2.0640
0.0173	7.0	525	0.2503	-4.7152	-18.9086	0.8800	14.1934	-228.4421	-67.0739	-1.9667	-2.0637
0.0866	7.3333	550	0.2469	-4.7026	-18.9134	0.8800	14.2107	-228.4901	-66.9480	-1.9667	-2.0638
0.0347	7.6667	575	0.2506	-4.7014	-18.9296	0.8800	14.2282	-228.6524	-66.9354	-1.9673	-2.0644
0.0866	8.0	600	0.2593	-4.7150	-18.9165	0.8800	14.2016	-228.5215	-67.0713	-1.9673	-2.0643
0.104	8.3333	625	0.2524	-4.7020	-18.9396	0.8800	14.2376	-228.7525	-66.9418	-1.9674	-2.0645
0.0173	8.6667	650	0.2503	-4.7003	-18.9479	0.8800	14.2476	-228.8348	-66.9245	-1.9674	-2.0645
0.0693	9.0	675	0.2511	-4.7024	-18.9458	0.8800	14.2433	-228.8138	-66.9459	-1.9682	-2.0653
0.0693	9.3333	700	0.2497	-4.7013	-18.9371	0.8800	14.2358	-228.7275	-66.9345	-1.9677	-2.0647
0.0693	9.6667	725	0.2438	-4.6908	-18.9453	0.8900	14.2545	-228.8092	-66.8294	-1.9677	-2.0648
0.0866	10.0	750	0.2475	-4.6928	-18.9485	0.8800	14.2557	-228.8416	-66.8499	-1.9679	-2.0651
0.0866	10.3333	775	0.2441	-4.7016	-18.9488	0.8900	14.2472	-228.8439	-66.9371	-1.9684	-2.0656
0.0866	10.6667	800	0.2484	-4.7051	-18.9371	0.8800	14.2320	-228.7267	-66.9721	-1.9685	-2.0656
0.0693	11.0	825	0.2421	-4.6951	-18.9478	0.8900	14.2526	-228.8337	-66.8728	-1.9685	-2.0656
0.052	11.3333	850	0.2506	-4.7064	-18.9338	0.8800	14.2275	-228.6942	-66.9851	-1.9684	-2.0655
0.0693	11.6667	875	0.2504	-4.6954	-18.9498	0.8800	14.2544	-228.8539	-66.8751	-1.9685	-2.0657
0.0693	12.0	900	0.2470	-4.6944	-18.9471	0.8800	14.2527	-228.8271	-66.8655	-1.9680	-2.0651
0.0347	12.3333	925	0.2470	-4.6947	-18.9470	0.8800	14.2523	-228.8264	-66.8687	-1.9680	-2.0651
0.0693	12.6667	950	0.2525	-4.6997	-18.9435	0.8800	14.2438	-228.7915	-66.9187	-1.9681	-2.0652
0.052	13.0	975	0.2525	-4.6991	-18.9443	0.8800	14.2452	-228.7995	-66.9127	-1.9681	-2.0652
0.0693	13.3333	1000	0.2525	-4.6992	-18.9442	0.8800	14.2450	-228.7986	-66.9138	-1.9681	-2.0652

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_1e6rate_01beta_CSFTDPO

Base model

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

Finetuned

tsavage68/UTI_M2_1000steps_1e7rate_SFT

Finetuned

(32)

this model