File size: 4,831 Bytes
0ae6d1e 2f7303b 0ae6d1e 8a1272d 0ae6d1e 4cd0001 2f7303b 0ae6d1e 2f7303b 4cd0001 0ae6d1e |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 |
---
license: apache-2.0
tags:
- moe
train: false
inference: false
pipeline_tag: text-generation
---
## Mixtral-8x7B-Instruct-v0.1-hf-attn-4bit-moe-3bit-metaoffload-HQQ
This is a version of the
<a href="https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1"> Mixtral-8x7B-Instruct-v0.1 model</a> quantized with a mix of 4-bit and 3-bit via Half-Quadratic Quantization (HQQ). More specifically, the attention layers are quantized to 4-bit and the experts are quantized to 3-bit.
Contrary to the <a href="https://huggingface.co/mobiuslabsgmbh/Mixtral-8x7B-Instruct-v0.1-hf-attn-4bit-moe-2bitgs8-metaoffload-HQQ"> 2bitgs8 model </a> that was designed to use less GPU memory, this one uses about ~22GB for the folks who want to get better quality and use the maximum VRAM available on 24GB GPUs. It reaches an impressive <b>71.10</b> LLM leaderboard score, not too far from the original model's <b>72.62</b>.
----------------------------------------------------------------------------------------------------------------------------------
</p>
## Performance
| Models | Mixtral Original | HQQ quantized |
|-------------------|------------------|------------------|
| Runtime VRAM | 94 GB | <b> 22.3 GB</b> |
| ARC (25-shot) | 70.22 | 69.62 |
| Hellaswag (10-shot)| 87.63 | 86.05 |
| MMLU (5-shot) | 71.16 | 69.46 |
| TruthfulQA-MC2 | 64.58 | 62.63 |
| Winogrande (5-shot)| 81.37 | 81.06 |
| GSM8K (5-shot)| 60.73 | 57.77 |
| Average| 72.62 | 71.10 |
### Basic Usage
To run the model, install the HQQ library from https://github.com/mobiusml/hqq and use it as follows:
``` Python
import transformers
from threading import Thread
model_id = 'mobiuslabsgmbh/Mixtral-8x7B-Instruct-v0.1-hf-attn-4bit-moe-3bit-metaoffload-HQQ'
#Load the model
from hqq.engine.hf import HQQModelForCausalLM, AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = HQQModelForCausalLM.from_quantized(model_id)
#Optional: set backend/compile
#You will need to install CUDA kernels apriori
# git clone https://github.com/mobiusml/hqq/
# cd hqq/kernels && python setup_cuda.py install
from hqq.core.quantize import *
HQQLinear.set_backend(HQQBackend.ATEN_BACKPROP)
def chat_processor(chat, max_new_tokens=100, do_sample=True):
tokenizer.use_default_system_prompt = False
streamer = transformers.TextIteratorStreamer(tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens=True)
generate_params = dict(
tokenizer("<s> [INST] " + chat + " [/INST] ", return_tensors="pt").to('cuda'),
streamer=streamer,
max_new_tokens=max_new_tokens,
do_sample=do_sample,
top_p=0.90,
top_k=50,
temperature= 0.6,
num_beams=1,
repetition_penalty=1.2,
)
t = Thread(target=model.generate, kwargs=generate_params)
t.start()
outputs = []
for text in streamer:
outputs.append(text)
print(text, end="", flush=True)
return outputs
################################################################################################
#Generation
outputs = chat_processor("How do I build a car?", max_new_tokens=1000, do_sample=False)
```
### Quantization
You can reproduce the model using the following quant configs:
``` Python
from hqq.engine.hf import HQQModelForCausalLM, AutoTokenizer
model_id = "mistralai/Mixtral-8x7B-Instruct-v0.1"
model = HQQModelForCausalLM.from_pretrained(model_id, use_auth_token=hf_auth, cache_dir=cache_path)
#Quantize params
from hqq.core.quantize import *
attn_prams = BaseQuantizeConfig(nbits=4, group_size=64, offload_meta=True)
experts_params = BaseQuantizeConfig(nbits=3, group_size=64, offload_meta=True)
zero_scale_group_size = 128
attn_prams['scale_quant_params']['group_size'] = zero_scale_group_size
attn_prams['zero_quant_params']['group_size'] = zero_scale_group_size
experts_params['scale_quant_params']['group_size'] = zero_scale_group_size
experts_params['zero_quant_params']['group_size'] = zero_scale_group_size
quant_config = {}
#Attention
quant_config['self_attn.q_proj'] = attn_prams
quant_config['self_attn.k_proj'] = attn_prams
quant_config['self_attn.v_proj'] = attn_prams
quant_config['self_attn.o_proj'] = attn_prams
#Experts
quant_config['block_sparse_moe.experts.w1'] = experts_params
quant_config['block_sparse_moe.experts.w2'] = experts_params
quant_config['block_sparse_moe.experts.w3'] = experts_params
#Quantize
model.quantize_model(quant_config=quant_config, compute_dtype=torch.float16);
model.eval();
```
|