Instructions to use tsavage68/Transaminitis_L3_1000steps_1e7rate_03beta_CSFTDPO with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use tsavage68/Transaminitis_L3_1000steps_1e7rate_03beta_CSFTDPO with Transformers:

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

pipe = pipeline("text-generation", model="tsavage68/Transaminitis_L3_1000steps_1e7rate_03beta_CSFTDPO")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("tsavage68/Transaminitis_L3_1000steps_1e7rate_03beta_CSFTDPO")
model = AutoModelForCausalLM.from_pretrained("tsavage68/Transaminitis_L3_1000steps_1e7rate_03beta_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/Transaminitis_L3_1000steps_1e7rate_03beta_CSFTDPO with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "tsavage68/Transaminitis_L3_1000steps_1e7rate_03beta_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/Transaminitis_L3_1000steps_1e7rate_03beta_CSFTDPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/tsavage68/Transaminitis_L3_1000steps_1e7rate_03beta_CSFTDPO

SGLang

How to use tsavage68/Transaminitis_L3_1000steps_1e7rate_03beta_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/Transaminitis_L3_1000steps_1e7rate_03beta_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/Transaminitis_L3_1000steps_1e7rate_03beta_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/Transaminitis_L3_1000steps_1e7rate_03beta_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/Transaminitis_L3_1000steps_1e7rate_03beta_CSFTDPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use tsavage68/Transaminitis_L3_1000steps_1e7rate_03beta_CSFTDPO with Docker Model Runner:
```
docker model run hf.co/tsavage68/Transaminitis_L3_1000steps_1e7rate_03beta_CSFTDPO
```

Transaminitis_L3_1000steps_1e7rate_03beta_CSFTDPO

This model is a fine-tuned version of tsavage68/Transaminitis_L3_1000rate_1e7_SFT on an unknown dataset. It achieves the following results on the evaluation set:

Loss: 0.1392
Rewards/chosen: 1.7582
Rewards/rejected: -2.4140
Rewards/accuracies: 0.9300
Rewards/margins: 4.1722
Logps/rejected: -26.6014
Logps/chosen: -12.6736
Logits/rejected: -1.0630
Logits/chosen: -1.0536

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
0.6949	0.2	25	0.6922	-0.0089	-0.0113	0.5600	0.0023	-18.5922	-18.5640	-1.0661	-1.0649
0.689	0.4	50	0.6902	-0.0743	-0.0808	0.5700	0.0065	-18.8241	-18.7820	-1.0662	-1.0650
0.6966	0.6	75	0.6990	-0.0262	-0.0204	0.4500	-0.0058	-18.6227	-18.6216	-1.0666	-1.0653
0.6628	0.8	100	0.7163	-0.1175	-0.1132	0.4600	-0.0042	-18.9322	-18.9257	-1.0693	-1.0680
0.7015	1.0	125	0.6776	-0.4612	-0.5510	0.5400	0.0898	-20.3914	-20.0715	-1.0730	-1.0717
0.6668	1.2	150	0.6800	0.0635	0.0084	0.4600	0.0551	-18.5267	-18.3224	-1.0700	-1.0688
0.6409	1.4	175	0.6361	-0.0704	-0.1936	0.7700	0.1232	-19.2001	-18.7690	-1.0737	-1.0723
0.6123	1.6	200	0.6074	0.0180	-0.1849	0.7300	0.2030	-19.1711	-18.4741	-1.0741	-1.0726
0.6043	1.8	225	0.5988	0.3189	0.0943	0.75	0.2245	-18.2403	-17.4714	-1.0739	-1.0724
0.5608	2.0	250	0.5625	0.4639	0.1602	0.8300	0.3037	-18.0205	-16.9878	-1.0739	-1.0723
0.5206	2.2	275	0.5084	0.5476	0.0971	0.8700	0.4504	-18.2310	-16.7090	-1.0751	-1.0732
0.4709	2.4	300	0.4614	0.7169	0.1230	0.8800	0.5939	-18.1448	-16.1447	-1.0753	-1.0732
0.433	2.6	325	0.4388	1.1217	0.3690	0.8500	0.7527	-17.3245	-14.7951	-1.0729	-1.0703
0.2802	2.8	350	0.3165	1.1549	-0.0896	0.9200	1.2445	-18.8533	-14.6845	-1.0737	-1.0701
0.3044	3.0	375	0.2744	1.3071	-0.1921	0.9000	1.4992	-19.1949	-14.1771	-1.0729	-1.0688
0.284	3.2	400	0.2314	1.5290	-0.3317	0.9100	1.8607	-19.6603	-13.4374	-1.0713	-1.0663
0.1771	3.4	425	0.1947	1.6285	-0.6031	0.9300	2.2317	-20.5652	-13.1057	-1.0703	-1.0645
0.2167	3.6	450	0.1840	1.5135	-0.9678	0.9200	2.4813	-21.7808	-13.4893	-1.0704	-1.0643
0.1395	3.8	475	0.1771	1.5622	-1.1304	0.9200	2.6926	-22.3226	-13.3267	-1.0702	-1.0636
0.2148	4.0	500	0.1874	1.6351	-1.2063	0.9300	2.8414	-22.5757	-13.0837	-1.0669	-1.0600
0.0286	4.2	525	0.1577	1.6211	-1.6079	0.9200	3.2289	-23.9142	-13.1307	-1.0665	-1.0588
0.1347	4.4	550	0.1587	1.6299	-1.7763	0.9200	3.4062	-24.4757	-13.1011	-1.0654	-1.0575
0.0575	4.6	575	0.1473	1.6935	-1.9790	0.9200	3.6725	-25.1514	-12.8892	-1.0648	-1.0564
0.158	4.8	600	0.1509	1.6992	-2.0296	0.9300	3.7288	-25.3200	-12.8703	-1.0651	-1.0566
0.0607	5.0	625	0.1475	1.6735	-2.1752	0.9400	3.8488	-25.8055	-12.9557	-1.0647	-1.0558
0.1122	5.2	650	0.1404	1.7217	-2.2066	0.9400	3.9283	-25.9101	-12.7953	-1.0637	-1.0549
0.1517	5.4	675	0.1448	1.7525	-2.2515	0.9400	4.0040	-26.0596	-12.6925	-1.0628	-1.0538
0.0413	5.6	700	0.1418	1.7609	-2.3002	0.9400	4.0611	-26.2220	-12.6647	-1.0635	-1.0545
0.0528	5.8	725	0.1358	1.7517	-2.3539	0.9200	4.1056	-26.4012	-12.6954	-1.0634	-1.0542
0.243	6.0	750	0.1356	1.7676	-2.3804	0.9300	4.1480	-26.4894	-12.6424	-1.0623	-1.0531
0.1309	6.2	775	0.1522	1.7393	-2.3923	0.9300	4.1315	-26.5290	-12.7367	-1.0632	-1.0540
0.1943	6.4	800	0.1425	1.7703	-2.3997	0.9300	4.1699	-26.5535	-12.6333	-1.0628	-1.0534
0.0754	6.6	825	0.1308	1.7709	-2.4143	0.9400	4.1852	-26.6025	-12.6314	-1.0623	-1.0529
0.0603	6.8	850	0.1401	1.7587	-2.4040	0.9300	4.1627	-26.5682	-12.6720	-1.0630	-1.0537
0.0603	7.0	875	0.1402	1.7651	-2.3976	0.9300	4.1627	-26.5466	-12.6505	-1.0636	-1.0543
0.046	7.2	900	0.1397	1.7519	-2.4254	0.9300	4.1773	-26.6393	-12.6945	-1.0631	-1.0538
0.2102	7.4	925	0.1390	1.7602	-2.4168	0.9300	4.1770	-26.6105	-12.6669	-1.0631	-1.0537
0.2116	7.6	950	0.1392	1.7582	-2.4140	0.9300	4.1722	-26.6014	-12.6736	-1.0630	-1.0536
0.094	7.8	975	0.1392	1.7582	-2.4140	0.9300	4.1722	-26.6014	-12.6736	-1.0630	-1.0536
0.1121	8.0	1000	0.1392	1.7582	-2.4140	0.9300	4.1722	-26.6014	-12.6736	-1.0630	-1.0536

Framework versions

Transformers 4.40.2
Pytorch 2.0.0+cu117
Datasets 2.19.1
Tokenizers 0.19.1

Downloads last month: 8

Safetensors

Model size

8B params

Tensor type

F16

Model tree for tsavage68/Transaminitis_L3_1000steps_1e7rate_03beta_CSFTDPO

Base model

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

Finetuned

tsavage68/Transaminitis_L3_1000rate_1e7_SFT

Finetuned

(17)

this model