Update private release-card draft v3

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+assets/thestage-edge-models-header.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -5,35 +5,52 @@ base_model:
 base_model_relation: quantized
 library_name: llama.cpp
 pipeline_tag: text-generation
+thumbnail: https://huggingface.co/TheStageAI/Qwen3.5-4B-GGUF/resolve/main/assets/thestage-edge-models-header.png
 tags:
   - gguf
   - llama.cpp
-  - quantization
+  - quantized
   - mixed-precision
+  - local-inference
+  - qwen3.5
 ---
 
-# Qwen3.5 4B — TheStageAI GGUF
+<p align="center">
+  <img src="./assets/thestage-edge-models-header.png" width="100%" alt="TheStageAI Edge Models — the right model at every memory budget">
+</p>
 
-Four optimized language-model GGUF files for text-only deployment with llama.cpp.
+<h1 align="center">Qwen3.5 4B — TheStageAI GGUF</h1>
 
-> **Recommended: M · 2.39 GB** — **78.86 MMLU-Pro**, 99.1% of the BF16 reference.
+<p align="center">
+  Four GGUF checkpoints for local inference with llama.cpp, from 1.52 GB to 4.49 GB.
+  <br>
+  <strong>Start with M — 2.39 GB and 98% of the BF16 instruction-strict IFEval score in our evaluation.</strong>
+</p>
+
+<p align="center">
+  <a href="https://github.com/TheStageAI/edge-lm"><img src="./assets/cta-edge-lm.svg" height="42" alt="Explore edge-lm on GitHub"></a>
+  &nbsp;
+  <a href="https://docs.thestage.ai/"><img src="./assets/cta-docs.svg" height="42" alt="Read TheStageAI documentation"></a>
+  &nbsp;
+  <a href="https://app.thestage.ai/"><img src="./assets/cta-platform.svg" height="42" alt="Open TheStageAI Platform"></a>
+</p>
 
 **Qwen 3.5 family:** [0.8B](https://huggingface.co/TheStageAI/Qwen3.5-0.8B-GGUF) · [2B](https://huggingface.co/TheStageAI/Qwen3.5-2B-GGUF) · **[4B](https://huggingface.co/TheStageAI/Qwen3.5-4B-GGUF)** · [9B](https://huggingface.co/TheStageAI/Qwen3.5-9B-GGUF)
 
-## Choose a checkpoint
+## Start here
 
-| Variant | Size | Best for | File |
+| Tier | Size | Use | Download |
 | --- | ---: | --- | --- |
-| XS | 1.52 GB | Minimum footprint · non-thinking only | [GGUF](./Qwen3.5-4B-XS-TS-Q3_K_S.gguf) |
-| S | 1.90 GB | Compact | [GGUF](./Qwen3.5-4B-S-TS-Q4_K_S.gguf) |
-| **M** | **2.39 GB** | **Recommended · Q4 reference** | [GGUF](./Qwen3.5-4B-M-TS-Q4_K_M.gguf) |
-| L | 4.49 GB | Maximum fidelity | [GGUF](./Qwen3.5-4B-L-TS-Q8_0.gguf) |
+| XS | 1.52 GB | Smallest | [GGUF](./Qwen3.5-4B-XS-TS-Q3_K_S.gguf) |
+| S | 1.90 GB | Adaptive compact | [GGUF](./Qwen3.5-4B-S-TS-Q4_K_S.gguf) |
+| **M** | **2.39 GB** | **Recommended · Fixed Q4 reference** | [GGUF](./Qwen3.5-4B-M-TS-Q4_K_M.gguf) |
+| L | 4.49 GB | Q8 fidelity reference | [GGUF](./Qwen3.5-4B-L-TS-Q8_0.gguf) |
 
 Sizes use decimal MB/GB. Exact byte counts are recorded in [`release-manifest.json`](./release-manifest.json).
 
-## Quickstart
+## Run with llama.cpp
 
-The command below selects the recommended **M** checkpoint by exact filename:
+This command selects the recommended **M** checkpoint by exact filename:
 
 ```bash
 llama-cli \
@@ -51,29 +68,35 @@ llama-cli \
 | **M** | 80.22 / 86.09 | 78.86 |
 | L | 81.70 / 87.05 | 79.59 |
 
-- **IFEval:** 541 prompts, native chat template, `enable_thinking=false`, temperature 0.
-- **MMLU-Pro:** complete 12,032-question runs only, native chat template, `enable_thinking=true`, temperature 1, top-p 0.95, 32,768-token output limit.
-- `—` means that no complete headline MMLU-Pro run is reported. Partial-subject accuracies are not used as release scores.
+Only complete 12,032-question MMLU-Pro runs are reported. `—` means that no complete release score is available; partial-subject results are not promoted to headline metrics.
 
 The benchmarks intentionally exercise different modes: IFEval measures non-thinking instruction and format adherence, while MMLU-Pro includes long sampled reasoning.
 
-## Reasoning mode
+> **Reasoning:** use S, M, or L for long-form reasoning. XS prioritizes minimum size; run Qwen XS with `--reasoning off`.
+
+<details>
+<summary><b>Evaluation protocol and XS reasoning note</b></summary>
+
+- **IFEval:** 541 prompts, native chat template, `enable_thinking=false`, temperature 0.
+- **MMLU-Pro:** 12,032 questions, native chat template, `enable_thinking=true`, temperature 1, top-p 0.95, 32,768-token output limit.
+- The headline BF16-retention percentage uses instruction-strict IFEval and is capped at 100%; the uncapped raw scores are shown in the table.
+
+XS prioritizes minimum footprint and ordinary chat or instruction following. For long-form reasoning, start with S, M, or L. Run Qwen XS with `--reasoning off`. In our long-thinking evaluation, XS generated substantially longer trajectories and reached the 32,768-token output limit more often than S. The XS MMLU-Pro cell is consequently marked `—`.
 
-**XS is not a supported reasoning operating point.** It is released for minimum-footprint, non-thinking chat and instruction following. Use S, M, or L when reasoning quality matters. In Qwen XS, long-thinking evaluation produced unstable trajectories, substantially longer generations, and frequent 32,768-token output-limit hits. Run XS with `--reasoning off`. The XS MMLU-Pro cell is therefore marked `—` (N/A).
+</details>
 
-## About this release
+## How these checkpoints were built
 
-This release starts from [`Qwen/Qwen3.5-4B`](https://huggingface.co/Qwen/Qwen3.5-4B) at revision [`851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a`](https://huggingface.co/Qwen/Qwen3.5-4B/tree/851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a) and applies TheStageAI's adaptive per-tensor precision assignment with tuned quantization scales.
+This release starts from [`Qwen/Qwen3.5-4B`](https://huggingface.co/Qwen/Qwen3.5-4B) at revision [`851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a`](https://huggingface.co/Qwen/Qwen3.5-4B/tree/851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a).
 
-- **XS:** minimum-footprint adaptive mixed precision for non-thinking use.
-- **S:** compact adaptive mixed precision.
-- **M:** uniform Q4 tensor assignment and reference operating point.
-- **L:** uniform Q8_0 tensor assignment for maximum fidelity.
+- **XS and S** use model-specific mixed-precision schedules selected for explicit size targets.
+- **M** keeps fixed Q4_K precision across quantized decoder tensors.
+- **L** keeps fixed Q8_0 precision as the high-fidelity reference.
 
-The recommendation is selected separately for each base model from the release evaluations.
+All four checkpoints receive scale tuning without changing their stored quantized codes or tensor layout. The recommended tier is selected independently for each base model from end-to-end evaluation.
 
 <details>
-<summary><b>Quantization labels and technical details</b></summary>
+<summary><b>File types, Hub selectors, and effective BPW</b></summary>
 
 | Variant | Hub selector | GGUF file type | Whole-file BPW |
 | --- | --- | --- | ---: |
@@ -82,19 +105,21 @@ The recommendation is selected separately for each base model from the release e
 | M | `Q4_K_M` | `MOSTLY_Q4_K_M` | 4.538 |
 | L | `Q8_0` | `MOSTLY_Q8_0` | 8.533 |
 
-XS and S are TheStage mixed-precision schedules. Their public Q-type suffixes provide an approximate size class for Hub discoverability; exact tensor-type inventories are recorded in [`release-manifest.json`](./release-manifest.json). M and L use uniform Q4_K and Q8_0 assignments for quantized decoder tensors.
+XS and S are TheStageAI mixed-precision schedules. Their public Q-type suffixes provide approximate Hub size classes for discoverability; they do not describe every tensor in the payload. M and L use fixed Q4_K and Q8_0 assignments for quantized decoder tensors.
+
+The exact tensor-type inventory and SHA-256 digest for every file are recorded in [`release-manifest.json`](./release-manifest.json). File size does not include KV cache or runtime buffers, which grow with context length. GGUF runtimes treat multimodal projectors as separate artifacts; this release contains the language-model files listed above.
 
 </details>
 
-## Limitations
+## TheStageAI edge stack
+
+These checkpoints provide a portable llama.cpp deployment path. For native compressed models on Apple Silicon, explore [edge-lm](https://github.com/TheStageAI/edge-lm). For custom compression, compilation, and deployment workflows, use the [TheStageAI Platform](https://app.thestage.ai/) and [documentation](https://docs.thestage.ai/).
 
-- These are language-model GGUF files for text-only deployment. Multimodal projector files are not included.
-- File size does not include KV cache or runtime buffers; total memory use grows with context length.
-- Small score reversals between BF16 and quantized rows should be read as evaluation variation, not as a claim that quantization improves the base model.
+Optimizing for a specific device, latency target, or memory budget? [Talk to the TheStageAI team →](https://app.thestage.ai/contact)
 
-## Provenance
+## Reproducibility
 
-The exact base revision, byte sizes, GGUF file types, whole-file BPW, tensor inventories, SHA-256 digests, held-out KL values, and evaluation IDs are recorded in [`release-manifest.json`](./release-manifest.json). Use its SHA-256 digests to verify downloaded files.
+The exact base revision, byte sizes, GGUF file types, whole-file BPW, tensor inventories, SHA-256 digests, held-out KL values, and evaluation IDs are recorded in [`release-manifest.json`](./release-manifest.json). Use its SHA-256 digests to verify downloaded files. Export and load gates used [llama.cpp revision `bec4772f6a2527d371557b5d2032641e5ff7619c`](https://github.com/ggml-org/llama.cpp/commit/bec4772f6a2527d371557b5d2032641e5ff7619c).
 
 ## License
 

release-manifest.json

@@ -36,7 +36,7 @@
   },
   "display_name": "Qwen3.5 4B",
   "family": "Qwen 3.5",
-  "generated_at": "2026-07-20T22:39:04.914041+00:00",
+  "generated_at": "2026-07-20T23:43:48.671876+00:00",
   "license": "apache-2.0",
   "model_key": "qwen3p5_4b",
   "reasoning_policy": {
@@ -48,8 +48,14 @@
     "xs": "non_thinking_only"
   },
   "recommended_product": "M",
+  "recommended_quality": {
+    "metric": "ifeval_instruction_strict",
+    "ratio_to_bf16": 0.9835616438356164,
+    "retention_percent_display": 98
+  },
   "repo_id": "TheStageAI/Qwen3.5-4B-GGUF",
-  "schema": "thestageai.gguf_release_manifest_v2",
+  "schema": "thestageai.gguf_release_manifest_v3",
+  "tested_llama_cpp_revision": "bec4772f6a2527d371557b5d2032641e5ff7619c",
   "variants": [
     {
       "benchmarks": {
@@ -81,7 +87,7 @@
       "lm_head_policy": "skip_tied",
       "output_embedding_mode": "tied_alias",
       "parameter_count": 4205751296,
-      "positioning": "Minimum footprint \u00b7 non-thinking only",
+      "positioning": "Smallest",
       "product": "XS",
       "reasoning_support": "non_thinking_only",
       "recommended": false,
@@ -126,7 +132,7 @@
       "lm_head_policy": "skip_tied",
       "output_embedding_mode": "tied_alias",
       "parameter_count": 4205751296,
-      "positioning": "Compact",
+      "positioning": "Adaptive compact",
       "product": "S",
       "reasoning_support": "supported",
       "recommended": false,
@@ -173,7 +179,7 @@
       "lm_head_policy": "skip_tied",
       "output_embedding_mode": "tied_alias",
       "parameter_count": 4205751296,
-      "positioning": "Q4 reference",
+      "positioning": "Fixed Q4 reference",
       "product": "M",
       "reasoning_support": "supported",
       "recommended": true,
@@ -216,7 +222,7 @@
       "lm_head_policy": "skip_tied",
       "output_embedding_mode": "tied_alias",
       "parameter_count": 4205751296,
-      "positioning": "Maximum fidelity",
+      "positioning": "Q8 fidelity reference",
       "product": "L",
       "reasoning_support": "supported",
       "recommended": false,