Upload README.md with huggingface_hub

README.md

@@ -0,0 +1,85 @@
+---
+language:
+- en
+library_name: mlx
+license: other
+license_name: qwen-community-1.0
+license_link: LICENSE
+pipeline_tag: image-text-to-text
+base_model: Qwen/Qwen3.8-Flash-Next
+tags:
+- mlx
+- jang
+- jang-4m
+- quantized
+- apple-silicon
+- vision
+- video
+- reasoning
+- agent
+- tool-use
+- qwen4_exp
+- moe
+- ngram-embedding
+- imatrix
+- awq
+---
+
+# JANGQ-AI/Qwen3.8-Flash-Next-JANG_4M
+
+**The recommended quality/size balance — median KL 0.0039 vs bf16 at 96.6 GiB (~73 GiB resident with the SSD-served table).**
+
+A JANG bundle of [Qwen/Qwen3.8-Flash-Next](https://huggingface.co/Qwen/Qwen3.8-Flash-Next)
+— the Qwen4-architecture preview: a 125B mixture-of-experts (512 experts,
+6B active) with a 51B hashed n-gram embedding, Gated DeltaNet + Qwen Sparse
+Attention hybrid layers, gated-residual streams, and vision+video towers —
+quantized for Apple Silicon / MLX. Text, image and video weights are all
+present in this exact bundle. Native multi-token-prediction head preserved (4-bit).
+
+> Runtime support is landing — this repo is private while the serving path
+> is finalized.
+
+## Quality (measured, 5,931 held-out positions vs bf16)
+
+![JANG ladder](./jang_ladder_wide.png)
+
+| Tier | Size | RAM w/ SSD-table | median KL | top-1 | top-5 | top-10 |
+|---|---|---|---|---|---|---|
+| JANG_1L | 60.4 GiB | ~41 GiB | 0.0366 | 86.7% | 97.6% | 98.8% |
+| JANG_2L | 65.9 GiB | ~48 GiB | 0.0249 | 87.8% | 98.2% | 99.1% |
+| JANG_4S | 72.4 GiB | ~53 GiB | 0.0161 | 89.7% | 98.6% | 99.3% |
+| **JANG_4M** | **96.6 GiB** | **~73 GiB** | **0.0039** | **94.4%** | **99.7%** | **99.9%** |
+| JANG_6S | 106.9 GiB | ~83 GiB | 0.0035 | 94.8% | 99.7% | 99.9% |
+
+Margin-conditioned flip curves are monotone-decreasing on every tier —
+quantization noise lives in the reference model's own uncertainty band, with
+zero disagreement at high-confidence positions on the upper tiers.
+
+## The n-gram table & memory
+
+The 51B n-gram embedding can be served directly from SSD by supporting
+runtimes; the "RAM w/ SSD-table" column above is the resident footprint in
+that mode.
+
+## What's in the bundle
+
+- **Vision + video:** the full vision tower and both image and video
+  preprocessors ship in this exact bundle — image-text-to-text and video
+  understanding work out of the box on supporting runtimes (image and video
+  token ids, mRoPE positions, and the merger are all present).
+- **Multi-token prediction:** the model's native MTP head is preserved (trained multi-step). Enables self-speculative decode on supporting runtimes.
+- **Thinking + agentic:** thinking mode on by default with three reasoning
+  efforts and preserved thinking history; Hermes-style tool calling; the
+  instruct preset gives direct non-thinking responses.
+- **Long context:** 262,144 tokens native, extensible to 1M with YaRN.
+
+## Serving contract
+
+- Thinking mode ON by default: `temperature=1.0, top_p=0.95, top_k=20`
+- Instruct mode: `temperature=0.7, top_p=0.80, top_k=20, presence_penalty=1.5`
+- Reasoning efforts `low / medium / xhigh` (default **xhigh**) and
+  `preserve_thinking` (default **on**) via chat-template kwargs
+- Context 262,144 native, extensible to 1M with YaRN
+- EOS `[248046, 248044]` · tool calls: Hermes-style `<tool_call>`
+
+Quantized and validated by **Jinho Jang** — eric@jangq.ai