Instructions to use tsavage68/UTI_L3_1000steps_1e6rate_05beta_CSFTDPO with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use tsavage68/UTI_L3_1000steps_1e6rate_05beta_CSFTDPO with Transformers:

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

pipe = pipeline("text-generation", model="tsavage68/UTI_L3_1000steps_1e6rate_05beta_CSFTDPO")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("tsavage68/UTI_L3_1000steps_1e6rate_05beta_CSFTDPO")
model = AutoModelForCausalLM.from_pretrained("tsavage68/UTI_L3_1000steps_1e6rate_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/UTI_L3_1000steps_1e6rate_05beta_CSFTDPO with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "tsavage68/UTI_L3_1000steps_1e6rate_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/UTI_L3_1000steps_1e6rate_05beta_CSFTDPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/tsavage68/UTI_L3_1000steps_1e6rate_05beta_CSFTDPO

SGLang

How to use tsavage68/UTI_L3_1000steps_1e6rate_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/UTI_L3_1000steps_1e6rate_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/UTI_L3_1000steps_1e6rate_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/UTI_L3_1000steps_1e6rate_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/UTI_L3_1000steps_1e6rate_05beta_CSFTDPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use tsavage68/UTI_L3_1000steps_1e6rate_05beta_CSFTDPO with Docker Model Runner:
```
docker model run hf.co/tsavage68/UTI_L3_1000steps_1e6rate_05beta_CSFTDPO
```

UTI_L3_1000steps_1e6rate_05beta_CSFTDPO

This model is a fine-tuned version of tsavage68/UTI_L3_1000steps_1e5rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:

Loss: 0.0078
Rewards/chosen: 2.5926
Rewards/rejected: -13.7164
Rewards/accuracies: 0.9900
Rewards/margins: 16.3089
Logps/rejected: -90.6274
Logps/chosen: -27.2939
Logits/rejected: -1.3641
Logits/chosen: -1.3371

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
0.5483	0.3333	25	0.2812	0.2629	-1.0915	0.9900	1.3544	-65.3776	-31.9532	-1.3242	-1.3091
0.0186	0.6667	50	0.0204	1.4814	-6.2620	0.9900	7.7434	-75.7187	-29.5163	-1.3323	-1.3150
0.0007	1.0	75	0.0124	2.1023	-8.8403	0.9900	10.9426	-80.8753	-28.2744	-1.3425	-1.3227
0.0174	1.3333	100	0.0110	2.7866	-9.0480	0.9900	11.8346	-81.2906	-26.9057	-1.3476	-1.3272
0.0173	1.6667	125	0.0107	2.2710	-10.8326	0.9900	13.1036	-84.8600	-27.9370	-1.3498	-1.3269
0.0348	2.0	150	0.0079	2.5738	-13.4526	0.9900	16.0264	-90.0999	-27.3315	-1.3620	-1.3349
0.0	2.3333	175	0.0079	2.5665	-13.4456	0.9900	16.0121	-90.0858	-27.3459	-1.3620	-1.3348
0.0	2.6667	200	0.0078	2.5714	-13.4484	0.9900	16.0198	-90.0914	-27.3362	-1.3619	-1.3348
0.0	3.0	225	0.0079	2.5744	-13.4805	0.9900	16.0549	-90.1557	-27.3302	-1.3623	-1.3352
0.0173	3.3333	250	0.0078	2.5790	-13.4989	0.9900	16.0779	-90.1926	-27.3210	-1.3623	-1.3352
0.0173	3.6667	275	0.0077	2.5749	-13.5072	0.9900	16.0821	-90.2091	-27.3291	-1.3623	-1.3351
0.0347	4.0	300	0.0078	2.5828	-13.5202	0.9900	16.1030	-90.2351	-27.3134	-1.3626	-1.3355
0.0	4.3333	325	0.0077	2.5858	-13.5544	0.9900	16.1403	-90.3036	-27.3074	-1.3626	-1.3355
0.0173	4.6667	350	0.0078	2.5816	-13.5650	0.9900	16.1466	-90.3246	-27.3158	-1.3628	-1.3357
0.0347	5.0	375	0.0079	2.5779	-13.5622	0.9900	16.1400	-90.3190	-27.3233	-1.3628	-1.3356
0.0173	5.3333	400	0.0077	2.5852	-13.5789	0.9900	16.1641	-90.3526	-27.3087	-1.3630	-1.3358
0.0347	5.6667	425	0.0078	2.5848	-13.6053	0.9900	16.1901	-90.4053	-27.3094	-1.3632	-1.3361
0.0173	6.0	450	0.0077	2.5855	-13.6105	0.9900	16.1960	-90.4156	-27.3079	-1.3634	-1.3364
0.0	6.3333	475	0.0079	2.5850	-13.6238	0.9900	16.2087	-90.4422	-27.3091	-1.3635	-1.3364
0.0347	6.6667	500	0.0077	2.5926	-13.6436	0.9900	16.2362	-90.4819	-27.2938	-1.3635	-1.3364
0.0	7.0	525	0.0077	2.5890	-13.6520	0.9900	16.2410	-90.4987	-27.3010	-1.3635	-1.3364
0.0	7.3333	550	0.0077	2.5868	-13.6463	0.9900	16.2331	-90.4873	-27.3054	-1.3636	-1.3365
0.0173	7.6667	575	0.0077	2.5918	-13.6721	0.9900	16.2639	-90.5389	-27.2955	-1.3637	-1.3366
0.0347	8.0	600	0.0078	2.5868	-13.6787	0.9900	16.2654	-90.5520	-27.3055	-1.3638	-1.3367
0.0347	8.3333	625	0.0077	2.5930	-13.6789	0.9900	16.2719	-90.5525	-27.2931	-1.3639	-1.3368
0.0	8.6667	650	0.0078	2.5892	-13.6871	0.9900	16.2763	-90.5689	-27.3006	-1.3638	-1.3367
0.0	9.0	675	0.0077	2.5903	-13.6943	0.9900	16.2847	-90.5834	-27.2984	-1.3639	-1.3368
0.0173	9.3333	700	0.0078	2.5860	-13.7028	0.9900	16.2888	-90.6002	-27.3070	-1.3642	-1.3371
0.0173	9.6667	725	0.0077	2.5865	-13.6964	0.9900	16.2830	-90.5876	-27.3060	-1.3641	-1.3370
0.0	10.0	750	0.0077	2.5939	-13.7066	0.9900	16.3006	-90.6079	-27.2912	-1.3641	-1.3370
0.0	10.3333	775	0.0079	2.5928	-13.7020	0.9900	16.2947	-90.5986	-27.2935	-1.3640	-1.3369
0.0173	10.6667	800	0.0078	2.5909	-13.7013	0.9900	16.2922	-90.5973	-27.2972	-1.3642	-1.3371
0.0173	11.0	825	0.0076	2.5913	-13.7123	0.9900	16.3036	-90.6193	-27.2965	-1.3641	-1.3370
0.0	11.3333	850	0.0077	2.5908	-13.7072	0.9900	16.2980	-90.6090	-27.2974	-1.3642	-1.3371
0.0347	11.6667	875	0.0078	2.5953	-13.7055	0.9900	16.3008	-90.6056	-27.2884	-1.3640	-1.3369
0.0	12.0	900	0.0078	2.5866	-13.7139	0.9900	16.3005	-90.6224	-27.3058	-1.3642	-1.3370
0.0173	12.3333	925	0.0077	2.5953	-13.6932	0.9900	16.2885	-90.5811	-27.2884	-1.3640	-1.3369
0.0173	12.6667	950	0.0077	2.5928	-13.7129	0.9900	16.3057	-90.6204	-27.2934	-1.3641	-1.3370
0.0347	13.0	975	0.0078	2.5926	-13.7164	0.9900	16.3089	-90.6274	-27.2939	-1.3641	-1.3371
0.0	13.3333	1000	0.0078	2.5926	-13.7164	0.9900	16.3089	-90.6274	-27.2939	-1.3641	-1.3371

Framework versions

Transformers 4.41.2
Pytorch 2.0.0+cu117
Datasets 2.19.2
Tokenizers 0.19.1

Downloads last month: 8

Safetensors

Model size

8B params

Tensor type

F16

Model tree for tsavage68/UTI_L3_1000steps_1e6rate_05beta_CSFTDPO

Base model

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

Finetuned

tsavage68/UTI_L3_1000steps_1e5rate_SFT

Finetuned

(39)

this model