forge: regenerate the model card

README.md

@@ -1,6 +1,7 @@
 ---
 license: apache-2.0
 license_link: https://huggingface.co/tencent/Hy3/blob/main/LICENSE
+thumbnail: https://huggingface.co/AtomicChat/Hy3-GGUF/resolve/main/hero.png
 base_model:
 - tencent/Hy3
 base_model_relation: quantized
@@ -9,14 +10,12 @@ pipeline_tag: text-generation
 library_name: gguf
 tags:
 - atomic-chat
-- hunyuan
-- tencent
 - hy3
-- moe
+- tencent
 - gguf
+- llama.cpp
 - imatrix
 - quantized
-- llama.cpp
 ---
 
 <center>
@@ -29,24 +28,21 @@ tags:
 
 <br/>
 
-<img src="https://huggingface.co/AtomicChat/Hy3-GGUF/resolve/main/hero.png" alt="Hy3" style="width:380px; max-width:100%; height:auto; margin-bottom:0.6em;"/>
+<img src="https://huggingface.co/AtomicChat/Hy3-GGUF/resolve/main/hero.png" alt="Hy3" style="width:100%; max-width:100%; height:auto; margin-bottom:0.6em;"/>
 
 <div style="display:flex; justify-content:center; gap:0.5em;">
 <a href="https://huggingface.co/tencent/Hy3"><strong>Base model: tencent/Hy3</strong></a>
 </div>
 </center>
 
-**Hy3** (Tencent Hunyuan Hy3), self-quantized to GGUF by [Atomic Chat](https://atomic.chat). Built straight from Tencent's original weights with a per-tensor importance matrix. Runs fully offline, including a 1-bit build that squeezes this 295B model onto a single multi-GPU box.
+**Hy3**, self-quantized to GGUF by [Atomic Chat](https://atomic.chat). Built straight from Tencent's original weights with a per-tensor importance matrix, so this is not a repack of somebody else's files. Runs fully offline.
 
 ## Highlights
 
-- **295B-parameter MoE with 21B active** (192 experts, top-8) plus a 3.8B MTP layer, so it delivers flagship-scale capability at a fraction of the active compute.
-- **Rivals open-source flagships with 2-5x its parameters**, per Tencent, and edges out GLM-5.1 in blind expert evaluation (2.67/4).
-- **Strong agentic and coding scores** (Tencent-reported): SWE-Bench Verified 78, GPQA Diamond 90.4, SWE-Bench Pro 57.9.
-- **Production-grade reliability**: stable tool calls, hallucination rate down to 5.4%, commonsense errors 12.7%, multi-turn context issues 7.9%.
-- **256K context** with grouped-query attention (8 KV heads, head dim 128) and an MTP layer for speculative decoding.
-- **Adjustable reasoning** via `reasoning_effort` (`no_think` / `low` / `high`) for chain-of-thought when a task needs it.
-- **Full imatrix quantization** over [`calibration_datav3`](https://gist.github.com/bartowski1182/eb213dccb3571f863da82e99418f81e8), including a 1-bit `IQ1_M`.
+- **298.8B parameters**: the weights this repo quantizes.
+- **Context length**: 262,144 tokens (256K), as published by Tencent.
+- **80 layers**: Mixture-of-Experts.
+- **Full imatrix ladder**: every quant is calibrated with an importance matrix, published here alongside the quants.
 
 > [!NOTE]
 > These GGUFs are **self-quantized from the original weights**, not a repack. The importance matrix keeps low-bit quants closer to the full-precision model.
@@ -59,24 +55,28 @@ tags:
 | Property | Value |
 |---|---|
 | Base model | `tencent/Hy3` |
-| Total / active parameters | 295B total / 21B active (plus a 3.8B MTP layer) |
-| Layers | 80 (plus 1 MTP layer) |
-| Experts | 192 experts, top-8 activated |
-| Context length | 256K |
-| Architecture | Mixture-of-Experts, GQA (8 KV heads, head dim 128), MTP speculative-decoding layer, reasoning modes |
-| This repo | GGUF quants (imatrix): `Q4_K_M` for near-reference quality, and a 1-bit `IQ1_M` (≈92GB) that makes this 295B model run locally. |
+| Parameters | 298.8B |
+| Layers | 80 |
+| Experts | 192 routed (top-8) |
+| Context length | 262,144 tokens (256K) |
+| Vocabulary | 120,832 |
+| Modalities | Text |
+| Architecture | Mixture-of-Experts, 192 experts (top-8), 64 attention heads over 8 KV heads, `HYV3ForCausalLM` |
+| This repo | GGUF quants (imatrix); the importance matrix is published here as `imatrix-atomic.gguf`. Quants: `IQ1_M`, `Q4_K_M` |
 
 <img src="https://huggingface.co/AtomicChat/Hy3-GGUF/resolve/main/benchmark.png" alt="Hy3 benchmark scores" style="width:100%; max-width:900px;"/>
 
-Scores are Tencent's published results for the base `tencent/Hy3`. Quantization preserves the large majority of this; `Q4_K_M` and up sit within a point or two of full precision.
-
+Scores are Tencent's published results for the base `tencent/Hy3`, not our own measurements. Quantization preserves the large majority of this; `Q4_K_M` and up stay close to full precision.
 
 ## Choosing a quant
 
 | Quant | Size | Notes |
 |---|---|---|
-| **`IQ1_M`** | 91.8 GB | **Smallest. 1-bit imatrix build that makes the 295B model run locally (≈92GB, e.g. a 4-GPU box). Expect quality tradeoffs; reasoning still works.** |
-| **`Q4_K_M`** | 184.7 GB | **Recommended for quality. Near-reference; needs roughly 185GB of combined VRAM and RAM.** |
+| `IQ1_M` | 91.8 GB | Last resort, only if nothing else fits. |
+| **`Q4_K_M`** | 184.7 GB | **Recommended default. Best balance of size, speed and quality.** |
+
+> [!TIP]
+> Pick the largest file that fits your (V)RAM with room for context. `Q4_K_M` is the sweet spot for most setups; `Q6_K` or `Q8_0` for maximum fidelity.
 
 ## Get started
 
@@ -93,31 +93,31 @@ Run Hy3 locally with:
 |---|---|
 | temperature | 0.9 |
 | top_p | 1.0 |
+| top_k | -1 |
 
-**Reasoning is recommended for Hy3.** Pass `reasoning_effort` (`high` for full chain-of-thought, `low`, or `no_think` for direct answers). On hard tasks, running without reasoning can produce truncated or malformed output.
+Tencent's recommended sampling configuration for `tencent/Hy3`.
 
 ## Run in llama.cpp
 
 ```bash
-git clone https://github.com/ggerganov/llama.cpp
+git clone https://github.com/ggml-org/llama.cpp
 cmake llama.cpp -B llama.cpp/build -DBUILD_SHARED_LIBS=OFF -DGGML_CUDA=ON
 cmake --build llama.cpp/build --config Release -j --target llama-cli llama-server
 ```
 
 ```bash
 ./llama.cpp/build/bin/llama-server \
-    -hf AtomicChat/Hy3-GGUF:IQ1_M \
+    -hf AtomicChat/Hy3-GGUF:Q4_K_M \
     --jinja -ngl 99 -c 8192 -fa on
 ```
 
 ## How these were made
 
 1. Download `tencent/Hy3` (original weights).
-2. Convert to GGUF with a [llama.cpp](https://github.com/ggerganov/llama.cpp) build that supports the Hy3 (`hy_v3`) architecture and its MTP layer.
-3. Build an importance matrix over `calibration_datav3`.
-4. Quantize with `--imatrix`: `Q4_K_M` for quality and `IQ1_M` for the smallest footprint that keeps this 295B model coherent.
+2. Convert to f16 GGUF with [llama.cpp](https://github.com/ggml-org/llama.cpp).
+3. Build an importance matrix over our calibration corpus, published here as `imatrix-atomic.gguf`.
+4. Quantize the ladder with `--imatrix`.
 
 ## License
 
-Released by Tencent under the Apache 2.0 license. Quantized by Atomic Chat.
-
+Original model by Tencent, released under the Apache 2.0 license. Full terms: [Apache 2.0](https://huggingface.co/tencent/Hy3/blob/main/LICENSE). Quantized by Atomic Chat.