Huihui Qwen3.8 27B Abliterated — Calibrated FP8

This repository contains a calibrated FP8 export of huihui-ai/Huihui-Qwen3.8-27B-abliterated with both FP8 model weights and FP8 scales for compressed cache/context.

Quantization

• Format: compressed-tensors 0.18.0, float-quantized
• Linear weights: symmetric FP8 (F8_E4M3) with static 128 × 128 weight blocks
• Input activations: symmetric dynamic FP8 with group size 128
• Observer: memoryless min-max
• Calibration: 512 samples with a maximum sequence length of 2,048 tokens
• Quantized weights: 256 linear weight tensors
• KV cache: static symmetric per-tensor FP8 scales calibrated for all 34 required scale tensors (32 main-model scales and 2 MTP scales)
• Retained in BF16: vision modules, linear-attention modules, token embeddings, language-model head, normalization tensors, and MTP tensors
• Protected-tensor verification: 943 retained BF16 tensors were verified bit-identical to the source checkpoint
• MTP: preserved as a separate BF16 checkpoint

The main converted checkpoint is split into four safetensors shards. The calibrated KV-cache scales are stored in kv_cache_scales.safetensors. Protected BF16 restoration tensors selected by the checkpoint index are stored in protected_bf16_restore.safetensors.

Uncensored model: the language checkpoint has undergone abliteration to reduce refusal behavior. Treat outputs as untrusted, apply application-level safeguards, and do not assume the model will decline harmful requests.

User responsibility: this model is provided without warranty. The creators, uploaders, and maintainers are not responsible or liable for what others generate, publish, deploy, or otherwise do with this abliterated model. Users must operate it responsibly, apply appropriate safeguards, comply with applicable law, and respect third-party rights. This model is for research purposes only and is not intended for production use.

The following is from the original Huihui model card:

huihui-ai/Huihui-Qwen3.8-27B-abliterated

This is an uncensored version of Qwen/Qwen3.8-27B created with abliteration (see remove-refusals-with-transformers to know more about it). This is a crude, proof-of-concept implementation to remove refusals from an LLM model without using TransformerLens.

Latest update

Only layers 18 to 51 have been ablated (Previously, The first 15 layers were retained without ablation), while the other layers remain unablated. This helps retain more of the original model’s performance. MTP and visual has not been modified.

If you want to download a previous version, just use the following command:

hf download huihui-ai/Huihui-Qwen3.8-27B-abliterated --local-dir ./huihui-ai/Huihui-Qwen3.8-27B-abliterated --revision d42ca89

Note

The first 15 layers were retained without ablation. MTP and visual has not been modified.

ollama

Please use the latest version of ollama

You can use huihui_ai/Qwen3.8-abliterated directly,

ollama run huihui_ai/Qwen3.8-abliterated

Usage

You can use this model in your applications by loading it with Hugging Face's transformers library:

#!/usr/bin/env python
# -*- coding: utf-8 -*-

import argparse
from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer
import torch
import os
import signal
import time

def parse_args():
    parser = argparse.ArgumentParser(
        description="Merge LoRA weights into huihui-ai/Huihui-Qwen3.8-27B-abliterated base model and save the full model."
    )
    parser.add_argument(
        "--base_model",
        type=str,
        default="huihui-ai/Huihui-Qwen3.8-27B-abliterated",
        help="HuggingFace repo or local path of the base model.",
    )
    parser.add_argument(
        "--dtype",
        type=str,
        default="bfloat16",
        choices=["float16", "bfloat16", "float32"],
        help="Data type for loading the base model (default: bfloat16).",
    )
    parser.add_argument(
        "--device_map",
        type=str,
        default="auto",
        help="Device map for model loading (e.g. 'cpu', 'auto').",
    )
    return parser.parse_args()

def main():
    cpu_count = os.cpu_count()
    print(f"Number of CPU cores in the system: {cpu_count}")
    half_cpu_count = cpu_count // 2
    os.environ["MKL_NUM_THREADS"] = str(half_cpu_count)
    os.environ["OMP_NUM_THREADS"] = str(half_cpu_count)
    torch.set_num_threads(half_cpu_count)

    print(f"PyTorch threads: {torch.get_num_threads()}")
    print(f"MKL threads: {os.getenv('MKL_NUM_THREADS')}")
    print(f"OMP threads: {os.getenv('OMP_NUM_THREADS')}")

    args = parse_args()

    # Load the model and tokenizer
    print(f"Load Model {args.base_model} ... ")

    torch_dtype = {
        "float16": torch.float16,
        "bfloat16": torch.bfloat16,
        "float32": torch.float32,
    }[args.dtype]

    model = AutoModelForCausalLM.from_pretrained(
        args.base_model,
        dtype=torch_dtype,
        device_map=args.device_map,
        trust_remote_code=True,
        low_cpu_mem_usage=True,
    )

    tokenizer = AutoTokenizer.from_pretrained(args.base_model, trust_remote_code=True)

    messages = []
    class CustomTextStreamer(TextStreamer):
        def __init__(self, tokenizer, skip_prompt=True, skip_special_tokens=True):
            super().__init__(tokenizer, skip_prompt=skip_prompt, skip_special_tokens=skip_special_tokens)
            self.generated_text = ""
            self.stop_flag = False
            self.init_time = time.time()  # Record initialization time
            self.end_time = None  # To store end time
            self.first_token_time = None  # To store first token generation time
            self.think_tokens_count = 0  # To track total think tokens
            self.token_count = 0  # To track total tokens

        def on_finalized_text(self, text: str, stream_end: bool = False):
            if self.first_token_time is None and text.strip():  # Set first token time on first non-empty text
                self.first_token_time = time.time()
            if stream_end:
                self.end_time = time.time()  # Record end time when streaming ends

            self.generated_text += text
            tokens = self.tokenizer.encode(text, add_special_tokens=False)
            self.token_count += len(tokens)
            if self.think_tokens_count == 0 and "</think>" in self.generated_text:
                self.think_tokens_count = self.token_count
            print(text, end="", flush=True)

            if self.stop_flag:
                raise StopIteration

        def stop_generation(self):
            self.stop_flag = True
            self.end_time = time.time()  # Record end time when generation is stopped

        def get_metrics(self):
            """Returns initialization time, first token time, first token latency, end time, total time, total tokens, and tokens per second."""
            if self.end_time is None:
                self.end_time = time.time()  # Set end time if not already set
            total_time = self.end_time - self.init_time  # Total time from init to end
            tokens_per_second = self.token_count / total_time if total_time > 0 else 0
            first_token_latency = (self.first_token_time - self.init_time) if self.first_token_time is not None else None
            metrics = {
                "init_time": self.init_time,
                "first_token_time": self.first_token_time,
                "first_token_latency": first_token_latency,
                "end_time": self.end_time,
                "total_time": total_time,  # Total time in seconds
                "total_tokens": self.token_count,
                "think_tokens_count": self.think_tokens_count,
                "real_tokens_count": self.token_count - self.think_tokens_count,
                "tokens_per_second": tokens_per_second
            }
            return metrics

    def generate_stream(model, tokenizer, messages, enable_thinking, skip_prompt, skip_special_tokens, max_new_tokens):
        text = tokenizer.apply_chat_template(
            messages,
            tokenize=False,
            add_generation_prompt=True,
            enable_thinking=enable_thinking
        )
        inputs = tokenizer(
            text,
            return_tensors="pt",
        ).to(model.device)

        streamer = CustomTextStreamer(tokenizer, skip_prompt=skip_prompt, skip_special_tokens=skip_special_tokens)

        def signal_handler(sig, frame):
            streamer.stop_generation()
            print("\n[Generation stopped by user with Ctrl+C]")

        signal.signal(signal.SIGINT, signal_handler)

        print("Response: ", end="", flush=True)
        try:
            generated_ids = model.generate(
                **inputs,
                max_new_tokens=max_new_tokens,
                streamer=streamer
            )
            del generated_ids
        except StopIteration:
            print("\n[Stopped by user]")

        del inputs
        torch.cuda.empty_cache()
        signal.signal(signal.SIGINT, signal.SIG_DFL)

        return streamer.generated_text, streamer.stop_flag, streamer.get_metrics()

    skip_prompt=True
    skip_special_tokens=True
    enable_thinking=False
    
    while True:
        print(f"skip_prompt = {skip_prompt}.")
        print(f"skip_special_tokens = {skip_special_tokens}.")
        print(f"enable_thinking = {enable_thinking}.")

        user_input = input("User: ").strip()
        if user_input.lower() == "/exit":
            print("Exiting chat.")
            break
        if user_input.lower() == "/clear":
            messages = []
            print("Chat history cleared. Starting a new conversation.")
            continue
        if user_input.lower() == "/skip_prompt":
            skip_prompt = not skip_prompt
            continue
        if user_input.lower() == "/skip_special_tokens":
            skip_special_tokens = not skip_special_tokens
            continue
        if user_input.lower() == "/enable_thinking":
            enable_thinking = not enable_thinking
            continue
        if not user_input:
            print("Input cannot be empty. Please enter something.")
            continue

        messages.append({"role": "user", "content": user_input})
        response, stop_flag, metrics = generate_stream(model, tokenizer, messages, enable_thinking, skip_prompt, skip_special_tokens, 40960)
        print("\n\nMetrics:")
        for key, value in metrics.items():
            print(f"  {key}: {value}")

        print("", flush=True)

        if stop_flag:
            continue
        messages.append({"role": "assistant", "content": response})

if __name__ == "__main__":
    main()

Usage Warnings

• Risk of Sensitive or Controversial Outputs: This model’s safety filtering has been significantly reduced, potentially generating sensitive, controversial, or inappropriate content. Users should exercise caution and rigorously review generated outputs.

• Not Suitable for All Audiences: Due to limited content filtering, the model’s outputs may be inappropriate for public settings, underage users, or applications requiring high security.

• Legal and Ethical Responsibilities: Users must ensure their usage complies with local laws and ethical standards. Generated content may carry legal or ethical risks, and users are solely responsible for any consequences.

• Research and Experimental Use: It is recommended to use this model for research, testing, or controlled environments, avoiding direct use in production or public-facing commercial applications.

• Monitoring and Review Recommendations: Users are strongly advised to monitor model outputs in real-time and conduct manual reviews when necessary to prevent the dissemination of inappropriate content.

• No Default Safety Guarantees: Unlike standard models, this model has not undergone rigorous safety optimization. huihui.ai bears no responsibility for any consequences arising from its use.

Donation

Your donation helps us continue our further development and improvement, a cup of coffee can do it.

• bitcoin:

  bc1qqnkhuchxw0zqjh2ku3lu4hq45hc6gy84uk70ge

• Support our work on Ko-fi!