speechless-sparsetral-mistral-16x7b-MoE

<p><h1> speechless-sparsetral-16x7b-MoE </h1></p>

speechless-sparsetral-16x7b-MoE is the MoE upgraded version of speechless-code-mistral-7b-v1.0. The MoE fine-tuning adopts Parameter-Efficient Sparsity Crafting (PESC), which is an efficient fine-tuning architecture that uses LoRA modules as expert models, similar to the concept of multi-loras. The model size is approximately 10B.

Specifically, Mistral-7B-0.1 is used as the base model, with 16 experts and 4 expert outputs selected for inference. The fine-tuning dataset includes codefuse-ai/Evol-Instruction-66k to enhance the model's code generation ability. The specific datasets are as follows:

• jondurbin/airoboros-2.2: Filter categories related to coding, reasoning and planning. 23,462 samples.
• Open-Orca/OpenOrca: Filter the 'cot' category in 1M GPT4 dataset. 74,440 samples.
• garage-bAInd/Open-Platypus: 100%, 24,926 samples.
• WizardLM/WizardLM_evol_instruct_V2_196k: Coding coversation part. 30,185 samples
• TokenBender/python_eval_instruct_51k: “python” in output .40,309 samples
• Spider: 8,659 samples
• codefuse-ai/Evol-Instruction-66k: 100%, 66,862 samples

Alpaca Prompt Format

### Instruction:
<instruction>
### Response:

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name_or_path="uukuguy/speechless-sparsetral-16x7b-MoE"
tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(model_name_or_path, device_map="auto", trust_remote_code=True).eval()

system = ""Below is an instruction that describes a task.\nWrite a response that appropriately completes the request.\n\n""
prompt = f"{system}\n\n### Instruction:\n{instruction}\n\n### Response:"

inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
pred = model.generate(**inputs, max_length=4096, do_sample=True, top_k=50, top_p=0.99, temperature=0.9, num_return_sequences=1)
print(tokenizer.decode(pred.cpu()[0], skip_special_tokens=True))