Blackfrost-AI

PINQWEN-3.6-9B-BF16

README

License: apache-2.0

Highlights

Reasons in the open. Native chain‑of‑thought via <think>…</think> — transparent, inspectable reasoning distilled from a panel of frontier teachers.
1,000,000‑token context. Full 1M context window (YaRN‑extended), backed by a gated‑linear‑attention hybrid backbone that keeps long‑context inference efficient.
Built for code and reasoning. Tuned for software engineering, step‑by‑step problem solving, and technical explanation.
Vision‑language. Retains a full vision encoder — text and image understanding in one 9B model.
Ships everywhere. Available in BF16, NVFP4, and a complete GGUF quant ladder (see Formats below).

This repository — BF16

Full‑precision (bfloat16) weights: the reference model, ideal for fine‑tuning, evaluation, vision‑language use, and as the source for further quantization. Retains the vision encoder and the multi‑token‑prediction head.

Quickstart

python
from transformers import AutoModelForImageTextToText, AutoProcessor

model = AutoModelForImageTextToText.from_pretrained(
    "Blackfrost-AI/PINQWEN-3.6-9B-BF16", dtype="bfloat16", device_map="auto")
proc = AutoProcessor.from_pretrained("Blackfrost-AI/PINQWEN-3.6-9B-BF16")

messages = [{"role": "user", "content": "Write a Python LRU cache with O(1) get/put and explain the design."}]
ids = proc.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)
out = model.generate(ids, max_new_tokens=1024)
print(proc.decode(out[0][ids.shape[1]:], skip_special_tokens=True))

Formats

Table with columns: Format, Repository, Best for
Format	Repository	Best for
BF16	`Blackfrost-AI/PINQWEN-3.6-9B-BF16`	Reference precision, fine‑tuning, vision
NVFP4	`Blackfrost-AI/PINQWEN-3.6-9B-NVFP4`	Fast serving on NVIDIA Blackwell
GGUF	`Blackfrost-AI/PINQWEN-3.6-9B-GGUF`	llama.cpp / local (full quant ladder, MTP)

Intended use

General reasoning, coding assistance, technical Q&A, long‑document and long‑codebase understanding, and multimodal (image + text) tasks. As with any language model, verify outputs for high‑stakes use.

License

Released under Apache‑2.0. Base architecture: Qwen 3.5 9B.

PINQWEN‑3.6‑9B is part of Blackfrost AI's Void model family.

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Model Details

Model Provider

Blackfrost-AI

Model Tree

Base

Qwen/Qwen3.5-9B

Fine-tuned

this model

Input Modalities

TextImageVideo

Output Modalities

Text

Supported Functionality