--- language: - en library_name: mlx license: apache-2.0 pipeline_tag: image-text-to-text base_model: Hcompany/Holo3-35B-A3B tags: - quantized - apple-silicon - mlx - qwen3_5_moe - moe - vision - hybrid-attention - gated-deltanet - turboquant - jangtq - jangtq2 - agent - computer-use - gui-agents ---

Holo3 35B-A3B — JANGTQ2 (MLX)

TurboQuant codebook quantization of H Company's Holo3 GUI-agent VLM — routed experts at 2-bit via Lloyd-Max codebooks + Hadamard rotation, attention / embed / shared-expert / lm_head at 8-bit affine, vision tower preserved.

--- ## Model Details | Property | Value | |---|---| | **Base model** | [`Hcompany/Holo3-35B-A3B`](https://huggingface.co/Hcompany/Holo3-35B-A3B) (finetune of `Qwen/Qwen3.5-35B-A3B`) | | **Parameters (source)** | 35 B total, ~3 B active per token | | **Architecture** | `qwen3_5_moe` — 40 decoder layers: 30 `Gated DeltaNet` (linear attn) + 10 full attention, 256 routed experts + 1 always-on shared expert | | **Quantization format** | `weight_format: mxtq` — routed experts via TurboQuant codebook (2-bit), everything else affine 8-bit or fp16 passthrough | | **Routed-expert storage** | `.tq_packed` (uint32) + `.tq_norms` (fp16) + `.tq_bits` (uint8); codebook + Hadamard signs re-derived deterministically at load | | **Package size on disk** | **11.63 GB** across 12 shards | | **Shipped tensors** | 1,930 total (1,597 language-model + 333 vision tower + 120 routed-expert TQ triples) | | **Vocab** | 248,320 | | **Context (position embeddings)** | 262,144 native | | **Vision tower** | 27-layer ViT (hidden 1152, patch 16), preserved in fp16 | | **Chat format** | Qwen `im_start`/`im_end` with `` reasoning toggle; Holo3 XML tool-call grammar | | **Use case** | GUI / computer-use agent (desktop, web, mobile) — designed for screenshot → action loops | ### Quantization details, per tensor category | Category | Bits | Group / codebook | Notes | |---|---|---|---| | **Routed-expert MLP** (`mlp.switch_mlp.gate_proj`, `up_proj`, `down_proj`) | **2 (JANGTQ)** | 2² Lloyd-Max centroids + Hadamard rotation | `.tq_packed` + `.tq_norms` + `.tq_bits` triples | | Embedding (`embed_tokens`), `lm_head` | 8 (affine) | group 64 | MLX-native `QuantizedLinear` | | Full-attention projections (`q_proj`, `k_proj`, `v_proj`, `o_proj`) | 8 (affine) | group 64 | Gate-doubled q_proj for `attn_output_gate` | | Linear-attention projections (`in_proj_qkv`, `in_proj_z`, `in_proj_b`, `in_proj_a`, `out_proj`) | 8 (affine) | group 64 | Gated DeltaNet | | Shared-expert MLP (`gate_proj`, `up_proj`, `down_proj`) | 8 (affine) | group 64 | Always active per token | | Router (`mlp.gate`) | fp16 passthrough | — | Precision-critical | | Shared-expert gate (`shared_expert_gate`) | fp16 passthrough | — | sigmoid scalar gate | | Norms (`*_layernorm`, `*_norm`), `A_log`, `dt_bias`, `conv1d` | fp16 passthrough | — | Un-quantized | | Vision tower (333 tensors) | fp16 passthrough | — | `patch_embed.proj` axes pre-transposed to MLX layout | JANGTQ ("TurboQuant") stores routed-expert weights as indices into a small Lloyd-Max codebook with a per-row norm, after a randomized Hadamard rotation that concentrates the distribution so quantization error is uniform. At inference, the input is rotated once per layer (cheap fused Metal kernel) and dot products happen against the codebook centroids directly, so we never dequantize back to affine. Compared to affine 2-bit at the same bit budget, this gives better quality **and** faster decode on the routed-expert MLP path. --- ## Usage **JANGTQ requires our custom loader** — stock `mlx_lm.load()` can't parse `.tq_packed` tensors. You need `jang-tools` (free, public): . ```bash pip install mlx mlx-lm mlx-vlm git clone https://github.com/jjang-ai/jangq && pip install -e ./jangq/jang-tools ``` ### Text ```python from jang_tools.load_jangtq import load_jangtq_model from mlx_lm import generate model, tokenizer = load_jangtq_model("JANGQ-AI/Holo3-35B-A3B-JANGTQ2") print(generate(model, tokenizer, prompt="The capital of France is", max_tokens=64)) ``` ### Image (VLM) — the intended use Holo3 is a GUI agent: give it a screenshot and it localizes UI elements and plans actions. ```python from jang_tools.load_jangtq_vlm import load_jangtq_vlm_model from mlx_vlm import generate from mlx_vlm.prompt_utils import apply_chat_template from mlx_vlm.utils import load_config path = "JANGQ-AI/Holo3-35B-A3B-JANGTQ2" model, processor = load_jangtq_vlm_model(path) config = load_config(path) prompt = apply_chat_template( processor, config, "Look at this desktop screenshot. Where should I click to open settings?", num_images=1, ) print(generate(model, processor, prompt, image="path/to/screenshot.png", max_tokens=256)) ``` ### Reasoning toggle ```python msgs = [{"role": "user", "content": "What is 17 × 23?"}] # Reasoning OFF — pre-closed block prompt = tokenizer.apply_chat_template(msgs, add_generation_prompt=True, enable_thinking=False) # Reasoning ON — model fills the block prompt = tokenizer.apply_chat_template(msgs, add_generation_prompt=True, enable_thinking=True) ``` Pass `enable_thinking` as a **direct kwarg** (the `chat_template_kwargs={...}` form only propagates on some tokenizer versions). ### Tool calls — Holo3 XML format Holo3 emits tool calls in a custom XML grammar (not JSON). Pass `tools=[...]` to the tokenizer's chat template; the model responds in this shape: ```xml 512 384 ``` Parse with a simple XML splitter on ``. See H Company's [quickstart](https://hub.hcompany.ai/quickstart) for a full agent harness example. ### Video The base model supports video via `transformers` and the bundle preserves `video_preprocessor_config.json`. `mlx-vlm` 0.4.4's `prepare_inputs` has no video path yet for `qwen3_5_moe` — for video, use upstream `transformers`. --- ## Hardware notes `11.63 GB` on disk; expect ~12–14 GB resident after load, plus KV cache. | Mac unified RAM | Works? | Notes | |---|---|---| | 16 GB | ✅ text-only | Image inference will be tight at long context | | 24 GB | ✅ comfortable | 32 k+ context, image inference OK | | 32 GB | ✅ | 100 k context viable, comfortable VL | | 64 GB+ | ✅ headroom | 262 k native context | --- ## Upstream benchmarks These are the base-model numbers for `Hcompany/Holo3-35B-A3B`, not evaluations of this JANGTQ2 quant: | Benchmark | Score | |---|---| | **OSWorld-Verified** (computer use) | **77.8 %** — SOTA at 3 B active | | WebArena (web navigation) | State-of-the-art (see upstream card) | | ScreenSpot-Pro (UI localization) | Top-tier (see upstream card) | | OSWorld-G (visual grounding) | Top-tier (see upstream card) | | H Corporate Benchmark (486 enterprise tasks) | Outperforms larger competitors | Independent JANGTQ-quant evaluation is tracked in the jang-tools repo and will land in future README revisions. --- ## Citation ```bibtex @misc{hai2025holo3modelfamily, title = {Holo3 - Open Foundation Models for Navigation and Computer Use Agents}, author = {H Company}, year = {2026}, url = {https://huggingface.co/Hcompany/Holo3-35B-A3B} } ``` ## License [Apache 2.0](https://huggingface.co/Hcompany/Holo3-35B-A3B/blob/main/LICENSE) — inherits from the base model. ---