Instructions to use OsaurusAI/Hy3-JANG_2K with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use OsaurusAI/Hy3-JANG_2K with MLX:

# Make sure mlx-lm is installed
# pip install --upgrade mlx-lm

# Generate text with mlx-lm
from mlx_lm import load, generate

model, tokenizer = load("OsaurusAI/Hy3-JANG_2K")

prompt = "Write a story about Einstein"
messages = [{"role": "user", "content": prompt}]
prompt = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True
)

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

Notebooks
Google Colab
Kaggle
Local Apps Settings
LM Studio

How to use OsaurusAI/Hy3-JANG_2K with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "OsaurusAI/Hy3-JANG_2K"

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "mlx-lm": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "OsaurusAI/Hy3-JANG_2K"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use OsaurusAI/Hy3-JANG_2K with Hermes Agent:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "OsaurusAI/Hy3-JANG_2K"

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default OsaurusAI/Hy3-JANG_2K

Run Hermes

hermes

OpenClaw new

How to use OsaurusAI/Hy3-JANG_2K with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "OsaurusAI/Hy3-JANG_2K"

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "OsaurusAI/Hy3-JANG_2K" \
  --custom-provider-id mlx-lm \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

MLX LM

How to use OsaurusAI/Hy3-JANG_2K with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "OsaurusAI/Hy3-JANG_2K"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "OsaurusAI/Hy3-JANG_2K"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "OsaurusAI/Hy3-JANG_2K",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

Hy3-JANG_2K / README.md

Osaurus-AI

Upload Hy3-JANG_2K via jang-tools

d08384c verified 3 days ago

preview code

Raw

History Blame Contribute Delete

4.67 kB

	---
	license: other
	license_name: tencent-hunyuan-community
	library_name: mlx
	base_model: tencent/Hy3
	base_model_relation: quantized
	pipeline_tag: text-generation
	tags:
	- hunyuan
	- hy3
	- mixture-of-experts
	- mlx
	- apple-silicon
	- reasoning
	- tool-use
	- quantized
	- jang
	- osaurus
	quantization_config:
	family: jang-affine-mixed
	profile: JANG_2K
	group_size: 128
	routed_avg_bits: 2.33
	---

	<p align="center"><img src="osaurus-x-banner.png" width="100%" alt="OsaurusAI"/></p>

	# Hy3-JANG_2K

	Quantized [tencent/Hy3](https://huggingface.co/tencent/Hy3) for Apple Silicon MLX / JANG runtimes — a 295B-total / 21B-active text MoE, packed to ~94 GiB. This is the clean non-MTP JANG_2K bundle (smallest 2K pack). For the variant that keeps Hy3's native Multi-Token-Prediction head, see `Hy3-JANG_2K-MTP`.

	\| \| \|
	\|---\|---\|
	\| Source \| [tencent/Hy3](https://huggingface.co/tencent/Hy3) \|
	\| License \| `other` — inherits the upstream Tencent Hunyuan Community License \|
	\| Architecture \| `hy_v3` (`HYV3ForCausalLM`), text-only \|
	\| Parameters \| 295B total / 21B active per token \|
	\| Format \| JANG_2K (mixed-affine), routed experts avg 2.33-bit \|
	\| Bundle size \| 101.40 GB (94.44 GiB), 22 shards, 2,876 tensor keys \|
	\| MTP \| none (`num_nextn_predict_layers = 0`) — MTP head not included \|
	\| Context \| 262,144 tokens \|

	## What this is

	`Hy3-JANG_2K` is a JANG mixed-affine quantization of Tencent's Hy3 dense-MoE, targeting Apple Silicon runtimes (MLX / vMLX). The `2K` profile spends an extra bit on the routed `down_proj` (3-bit vs the 2-bit `gate`/`up`), which cleans up the sampling tail relative to a uniform 2-bit pack. This bundle drops the native MTP layer for the smallest footprint; use `Hy3-JANG_2K-MTP` if you want speculative decoding.

	## Quantization (JANG_2K)

	\| Tensor family \| Policy \|
	\|---\|---\|
	\| Routed expert `gate_proj` / `up_proj` \| affine 2-bit, group size 128 \|
	\| Routed expert `down_proj` \| affine 3-bit, group size 128 \|
	\| Attention `q/k/v/o` \| affine 8-bit \|
	\| Shared expert \| affine 8-bit \|
	\| Dense layer-0 MLP \| affine 8-bit \|
	\| `embed_tokens` \| affine 6-bit \|
	\| `lm_head` \| affine 8-bit \|
	\| RMSNorms, router gate, expert bias \| 16-bit passthrough \|

	Routed-expert effective average: 2.33 bit. AWQ scaling is disabled for this bundle (measured negligible on Hy3).

	## Architecture

	Hy3 is a text-only dense-causal-GQA MoE — not MLA, not SSM, not sliding-window, not a VLM.

	- 80 decoder layers, `hidden_size` 4096
	- GQA: 64 attention heads / 8 KV heads, `head_dim` 128, QK-norm
	- RoPE `default`, `rope_theta` 11,158,840, `max_position_embeddings` 262,144
	- MoE: 192 routed experts, top-8, sigmoid router + expert-correction bias, `route_norm`, `router_scaling_factor` 2.826, 1 shared expert, `first_k_dense_replace` 1
	- No MTP layer in this bundle (`num_nextn_predict_layers = 0`)
	- `vocab_size` 120,832

	## Reasoning & tool use

	- Reasoning: `<think>…</think>` tags, `reasoning_effort` (`no_think` / `low` / `high`).
	- Tool calling: Hunyuan / Tencent XML-style tags (`<tool_calls>`, `<tool_call>`, `<arg_key>`, `<arg_value>`).
	- Hy3's tokenizer uses a `:opensource` special-token dialect (e.g. `<｜hy_eos:opensource｜>`, `<think:opensource>`); the bundled `chat_template.jinja` is the upstream template. A compatible runtime must resolve these variant-suffixed tokens at the token→text boundary.

	## Runtime support

	- Converted and structurally verified (index complete, 2,876 tensors / 22 shards, no MTP tensors).
	- Runs on the vMLX Python engine with Hy3 support: JANG affine loader, GQA KV cache, `<think>` reasoning stream, and Hunyuan tool-call parsing.

	Requires a Hy3-aware MLX/JANG runtime. Stock `mlx-lm` / `transformers` will not load the JANG mixed-affine layout as-is.

	## Known limitations

	- No published quality benchmark yet for this specific pack.
	- Very loose sampling (`top_p` 1.0 + `temperature` 0.9) exposes more of the routed-expert tail; a mild `top_p ≤ 0.9` or `min_p` floor is recommended for long-form generation.

	## 소개 (Korean)

	이 번들은 Tencent의 Hy3 (295B 총 파라미터 / 21B 활성 MoE, 텍스트 전용)를 Apple Silicon MLX / JANG 런타임용으로 양자화한 모델입니다. JANG_2K 프로파일은 라우팅 전문가의 `down_proj`를 3-bit로, `gate`/`up`을 2-bit로 양자화합니다(평균 2.33-bit). 이 번들은 MTP 헤드를 포함하지 않는 가장 작은 2K 팩이며, 스펙티브 디코딩이 필요하면 `Hy3-JANG_2K-MTP`를 사용하세요. Hy3의 GQA 어텐션과 MoE 라우팅, `:opensource` 특수 토큰 방식을 정확히 구현한 런타임에서만 사용해야 합니다.