Not-For-All-Audiences

Instructions to use DS-Archive/mistral-v0.1-7b-pippa-metharme-lora with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use DS-Archive/mistral-v0.1-7b-pippa-metharme-lora with Transformers:

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

pipe = pipeline("text-generation", model="DS-Archive/mistral-v0.1-7b-pippa-metharme-lora")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("DS-Archive/mistral-v0.1-7b-pippa-metharme-lora")
model = AutoModelForCausalLM.from_pretrained("DS-Archive/mistral-v0.1-7b-pippa-metharme-lora")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use DS-Archive/mistral-v0.1-7b-pippa-metharme-lora with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "DS-Archive/mistral-v0.1-7b-pippa-metharme-lora"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "DS-Archive/mistral-v0.1-7b-pippa-metharme-lora",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/DS-Archive/mistral-v0.1-7b-pippa-metharme-lora

SGLang

How to use DS-Archive/mistral-v0.1-7b-pippa-metharme-lora 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 "DS-Archive/mistral-v0.1-7b-pippa-metharme-lora" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "DS-Archive/mistral-v0.1-7b-pippa-metharme-lora",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "DS-Archive/mistral-v0.1-7b-pippa-metharme-lora" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "DS-Archive/mistral-v0.1-7b-pippa-metharme-lora",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use DS-Archive/mistral-v0.1-7b-pippa-metharme-lora with Docker Model Runner:
```
docker model run hf.co/DS-Archive/mistral-v0.1-7b-pippa-metharme-lora
```

mistral-v0.1-7b-pippa-metharme-lora

This model is a fine-tuned version of mistralai/Mistral-7B-v0.1 on the PIPPA dataset. It achieves the following results on the evaluation set:

Loss: 1.3494

Model description

8-bit lora trained on the PygmalionAI/PIPPA dataset using axolotl.

Intended uses & limitations

PIPPA consists of just a little more than 1 million lines of dialogue spread out over 26,000 conversations between users of the popular chatbot website "Character.AI" and its large language model, obtained through a large community effort taking place over the course of several months. Tallying shows that over 1,000 unique personas simulating both real and fictional characters are represented within the dataset, allowing PIPPA and LLMs fine-tuned on it to adapt to many different roleplay domains.

⚠️ CAUTION: PIPPA contains conversations, themes and scenarios which can be considered "not safe for work" (NSFW) and/or heavily disturbing in nature. Models trained purely with PIPPA may have the tendency to generate X-rated output. You have been warned.

Training and evaluation data

PygmalionAI/PIPPA

@misc{gosling2023pippa,
      title={PIPPA: A Partially Synthetic Conversational Dataset}, 
      author={Tear Gosling and Alpin Dale and Yinhe Zheng},
      year={2023},
      eprint={2308.05884},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.0001
train_batch_size: 2
eval_batch_size: 2
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 40
num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
1.7313	0.05	100	1.7044
1.68	0.11	200	1.6176
1.5642	0.16	300	1.5538
1.6617	0.22	400	1.4986
1.4733	0.27	500	1.4723
1.4916	0.33	600	1.4427
1.5036	0.38	700	1.4271
1.2385	0.44	800	1.4109
1.4094	0.49	900	1.3968
1.4042	0.55	1000	1.3848
1.3946	0.6	1100	1.3771
1.2523	0.66	1200	1.3692
1.2932	0.71	1300	1.3648
1.346	0.77	1400	1.3609
1.1163	0.82	1500	1.3565
1.4656	0.88	1600	1.3495
1.2698	0.93	1700	1.3484
1.2019	0.99	1800	1.3454
1.3685	1.04	1900	1.3477
1.2248	1.1	2000	1.3488
1.2162	1.15	2100	1.3479
1.0443	1.21	2200	1.3491
1.2445	1.26	2300	1.3460
1.3229	1.32	2400	1.3476
1.3464	1.37	2500	1.3439
1.2651	1.43	2600	1.3439
1.516	1.48	2700	1.3424
1.4323	1.54	2800	1.3413
1.08	1.59	2900	1.3436
1.289	1.64	3000	1.3379
1.1221	1.7	3100	1.3384
1.1895	1.75	3200	1.3376
1.3138	1.81	3300	1.3358
1.3907	1.86	3400	1.3343
1.4544	1.92	3500	1.3351
1.25	1.97	3600	1.3334
1.2682	2.03	3700	1.3452
1.3107	2.08	3800	1.3471
1.2096	2.14	3900	1.3496
1.4503	2.19	4000	1.3503
1.142	2.25	4100	1.3485
0.8439	2.3	4200	1.3490
1.2749	2.36	4300	1.3508
0.9578	2.41	4400	1.3502
1.2203	2.47	4500	1.3496
0.9451	2.52	4600	1.3498
0.9602	2.58	4700	1.3491
0.9501	2.63	4800	1.3491
1.2062	2.69	4900	1.3496
1.1728	2.74	5000	1.3491
1.2506	2.8	5100	1.3494
1.4052	2.85	5200	1.3494
1.2012	2.91	5300	1.3494
1.3141	2.96	5400	1.3494

Framework versions

Transformers 4.34.0.dev0
Pytorch 2.0.1+cu118
Datasets 2.14.5
Tokenizers 0.14.0

Downloads last month: 2

Model tree for DS-Archive/mistral-v0.1-7b-pippa-metharme-lora

Base model

mistralai/Mistral-7B-v0.1

Quantized

(193)

this model

Dataset used to train DS-Archive/mistral-v0.1-7b-pippa-metharme-lora

Paper for DS-Archive/mistral-v0.1-7b-pippa-metharme-lora

PIPPA: A Partially Synthetic Conversational Dataset

Paper • 2308.05884 • Published Aug 11, 2023 • 34