chore(qwen38-flash-next): move to the Unsloth Dynamic 3.1 UD-Q4_K_XL weights (#1031)

README.md

@@ -22,7 +22,7 @@ All model documentation lives under [docs/models](docs/models/README.md):
 | Model | Inference modes | Hugging Face weights | Benchmarks | Quality |
 | --- | --- | --- | --- | --- |
 | [Qwen3.8 27B](docs/models/qwen3.8-27b/README.md) | Q4/Q8, images, AR, DFlash2 | Unsloth [Q4_K_XL](https://huggingface.co/unsloth/Qwen3.8-27B-GGUF/blob/4ca720788d1e01f1bff70c033e0d0028fd02e502/Qwen3.8-27B-UD-Q4_K_XL.gguf) / [Q8_K_XL](https://huggingface.co/unsloth/Qwen3.8-27B-GGUF/blob/4ca720788d1e01f1bff70c033e0d0028fd02e502/Qwen3.8-27B-UD-Q8_K_XL.gguf) · [DFlash2 Q4_K_M](https://huggingface.co/z-lab/Qwen3.8-27B-DFlash2-GGUF/blob/2d9571f8ce46e151f61c6499c99dee6079e1d610/Qwen3.8-27B-DFlash2-Q4_K_M.gguf) | Q4: **648.89 tok/s pp**; up to **71.55 tok/s tg** single user and **128.42 aggregated tok/s** on 8 concurrent requests with DFlash2 · [Benchmarks](docs/models/qwen3.8-27b/BENCHMARKS.md) | [Quality](docs/models/qwen3.8-27b/QUALITY.md) |
-| [Qwen3.8 Flash-Next](docs/models/qwen3.8-flash-next/README.md) | Q4, images, AR, MTP | Unsloth [Q4_K_XL](https://huggingface.co/unsloth/Qwen3.8-Flash-Next-GGUF/tree/38bb39ee97821de2c9009abb7e93950eec396e66/UD-Q4_K_XL) · [MTP Q8_0](https://huggingface.co/unsloth/Qwen3.8-Flash-Next-GGUF/blob/38bb39ee97821de2c9009abb7e93950eec396e66/MTP/mtp-Qwen3.8-Flash-Next-shared-Q8_0.gguf) | **1,801.01 tok/s pp** (AR); up to **116.93 tok/s tg** single user and **245.81 aggregated tok/s** on 8 concurrent requests with MTP · [Benchmarks](docs/models/qwen3.8-flash-next/BENCHMARKS.md) | [Quality](docs/models/qwen3.8-flash-next/QUALITY.md) |
+| [Qwen3.8 Flash-Next](docs/models/qwen3.8-flash-next/README.md) | Q4, images, AR, MTP | Unsloth [Q4_K_XL](https://huggingface.co/unsloth/Qwen3.8-Flash-Next-GGUF/tree/d96201194f103c906994f283f504209dc8d2a10c/UD-Q4_K_XL) · [MTP Q8_0](https://huggingface.co/unsloth/Qwen3.8-Flash-Next-GGUF/blob/d96201194f103c906994f283f504209dc8d2a10c/MTP/mtp-Qwen3.8-Flash-Next-shared-Q8_0.gguf) | **1,801.01 tok/s pp** (AR); up to **116.93 tok/s tg** single user and **245.81 aggregated tok/s** on 8 concurrent requests with MTP · [Benchmarks](docs/models/qwen3.8-flash-next/BENCHMARKS.md) | [Quality](docs/models/qwen3.8-flash-next/QUALITY.md) |
 | [DeepSeek V4 Flash](docs/models/deepseek-v4-flash/README.md) | AR, DSpark | [antirez Flash 0731 IQ2XXS](https://huggingface.co/antirez/deepseek-v4-gguf/blob/1cd7b564460821938add0475a60b942c409295e0/DeepSeek-V4-Flash-IQ2XXS-w2Q2K-AProjQ8-SExpQ8-OutQ8-chat-v2-imatrix-0731.gguf) · [DSpark](https://huggingface.co/antirez/deepseek-v4-gguf/blob/e7f04037032990db0346398d249baf9fb9df1ccc/DeepSeek-V4-Flash-DSpark-support-0731.gguf) | **474.87 tok/s pp**; up to **29.57 tok/s tg** single user with DSpark and **77.12 aggregated tok/s** on 8 concurrent requests with AR · [Benchmarks](docs/models/deepseek-v4-flash/BENCHMARKS.md) | [Quality](docs/models/deepseek-v4-flash/QUALITY.md) |
 | [Qwen3-ASR 1.7B](docs/models/qwen3-asr/README.md) | Speech recognition | [BF16](https://huggingface.co/Qwen/Qwen3-ASR-1.7B/tree/7278e1e70fe206f11671096ffdd38061171dd6e5) | **15.27× realtime** · [Benchmarks](docs/models/qwen3-asr/BENCHMARKS.md) | [Quality](docs/models/qwen3-asr/QUALITY.md) |
 | [EmbeddingGemma 2](docs/models/embeddinggemma-2/README.md) | Text and image embeddings | [BF16](https://huggingface.co/google/embeddinggemma-2/tree/914f7f89142e33e77833254d9c9b90c3cef7303b) | Up to **41.5k tokens/s** over HTTP; **13.0 ms** for a 404-token request · [Benchmarks](docs/models/embeddinggemma-2/BENCHMARKS.md) | [Quality](docs/models/embeddinggemma-2/QUALITY.md) |

docs/models/qwen3.8-flash-next/README.md

@@ -1,7 +1,10 @@
 # Qwen3.8 Flash-Next
 
 Hybrid recurrent/QSA mixture-of-experts text/image model on gfx1151.
-Supported target: `unsloth/Qwen3.8-Flash-Next-GGUF`, **UD-Q4_K_XL** (four shards).
+Supported target: `unsloth/Qwen3.8-Flash-Next-GGUF`, **UD-Q4_K_XL** (four shards),
+Unsloth Dynamic 3.1 revision `d96201194f10`. It keeps the tensor types and
+shapes of the earlier `38bb39ee9782`, which the retained benchmark, quality
+and F32-fixture evidence used; AR pp2048/tg128 speed is unchanged.
 Optional shared-Q8 MTP predictor; optional BF16 vision projector.
 Original unquantized-model and GGUF-conversion parity remain unqualified.
 
@@ -11,8 +14,9 @@ Original unquantized-model and GGUF-conversion parity remain unqualified.
 
 ```sh
 nix develop -c hf download unsloth/Qwen3.8-Flash-Next-GGUF \
-  --revision 38bb39ee97821de2c9009abb7e93950eec396e66 \
-  --include "UD-Q4_K_XL/*" "MTP/mtp-Qwen3.8-Flash-Next-shared-Q8_0.gguf" "mmproj-BF16.gguf" \
+  --revision d96201194f103c906994f283f504209dc8d2a10c \
+  --include "UD-Q4_K_XL/*" --include "MTP/mtp-Qwen3.8-Flash-Next-shared-Q8_0.gguf" \
+  --include "mmproj-BF16.gguf" \
   --local-dir models/qwen3.8-flash-next
 nix build
 MODEL=/path/to/first-target-shard.gguf

src/models/qwen38_flash_next/functional/devlab.json

@@ -2,7 +2,7 @@
   "assets": {
     "flash-next": {
       "repo": "unsloth/Qwen3.8-Flash-Next-GGUF",
-      "revision": "38bb39ee97821de2c9009abb7e93950eec396e66",
+      "revision": "d96201194f103c906994f283f504209dc8d2a10c",
       "patterns": [
         "UD-Q4_K_XL/*.gguf",
         "MTP/mtp-Qwen3.8-Flash-Next-shared-Q8_0.gguf",

tests/functional/devlab.json

@@ -12,7 +12,7 @@
   "assets": {
     "flash-next": {
       "repo": "unsloth/Qwen3.8-Flash-Next-GGUF",
-      "revision": "38bb39ee97821de2c9009abb7e93950eec396e66",
+      "revision": "d96201194f103c906994f283f504209dc8d2a10c",
       "patterns": [
         "UD-Q4_K_XL/*.gguf",
         "MTP/mtp-Qwen3.8-Flash-Next-shared-Q8_0.gguf",

tests/models/qwen38_flash_next/gpu_probe.cpp

@@ -415,8 +415,7 @@ int main(int argc, char** argv) {
     if (!executor->Forward(
             *session, std::span<const std::int32_t>(tokens.data() + off, n),
             rows, out.data(), q::rocm::Executor::ForwardMode::kPrefill, &error,
-            nullptr,
-            std::span<const std::int32_t>(tokens.data() + off + n, next))) {
+            {}, std::span<const std::int32_t>(tokens.data() + off + n, next))) {
       std::fprintf(stderr, "prefill failed: %s\n", error.c_str());
       return 1;
     }