kernelpool's picture
Update README.md
806abdf verified
metadata
license: apache-2.0
pipeline_tag: text-generation
library_name: mlx
tags:
  - mlx
base_model: ByteDance-Seed/Seed-OSS-36B-Instruct
language:
  - en
  - zh

catalystsec/Seed-OSS-36B-Instruct-4bit-DWQ

This model was quantized to 4-bit using DWQ with mlx-lm version 0.27.1, distilled from a BF16 teacher model.

Learning Rate Total Loss KL Loss Activation Loss Improvement
2e-7 0.415 0.025 0.390 15.8%

Use with mlx

pip install mlx-lm
from mlx_lm import load, generate

model, tokenizer = load("catalystsec/Seed-OSS-36B-Instruct-4bit-DWQ")

prompt = "hello"

if tokenizer.chat_template is not None:
    messages = [{"role": "user", "content": prompt}]
    prompt = tokenizer.apply_chat_template(
        messages, add_generation_prompt=True
    )

response = generate(model, tokenizer, prompt=prompt, verbose=True)