{"slug": "dwarf-star-support-for-qwen3-8-flash-next", "title": "Dwarf Star Support for Qwen3.8 Flash Next", "summary": "Developer ivanfioravanti merged a pull request adding Qwen3.8 Flash Next support to the antirez/ds4 repository, including qwen4exp loading, tokenizer and chat formatting, PLE sidecars, Metal text and vision graphs, built-in MTP, and recurrent KV persistence. The branch was force-pushed multiple times between September 6 and September 13, 2026, and a contributor flagged that the Q4_1 encoder from commit a0b1658 was dropped in b020c9f and the squashed 3030554 while qwen4_pack.py still uses it for the PLE sidecar. A Q4_K imatrix test model was published at huggingface.co/ivanfioravanti/Qwen3.8-Flash-Next-DS4-Q4, and the PR includes reproducible throughput charts and measurements on M3 Ultra.", "body_md": "# Qwen3.8 flash next - #991\n\nMerged\n\nMerged\n\n## Conversation\n\nContributor\nAuthor\n\n| Q4_K imatrix model for testing is [here](https://huggingface.co/ivanfioravanti/Qwen3.8-Flash-Next-DS4-Q4) | \n\n[ivanfioravanti](https://github.com/ivanfioravanti)\n\n[force-pushed](https://github.com/antirez/ds4/compare/ecf1da3c9516d0d78a31c61ba8cba8b535488caa..236cb2a549d05ea1941a19bd4f154e0abc858661)the qwen3.8-flash-next branch from\n\n[to](https://github.com/antirez/ds4/commit/ecf1da3c9516d0d78a31c61ba8cba8b535488caa)\n\n`ecf1da3`\n\n`236cb2a`\n[Compare](https://github.com/antirez/ds4/compare/ecf1da3c9516d0d78a31c61ba8cba8b535488caa..236cb2a549d05ea1941a19bd4f154e0abc858661)\n\n[September 6, 2026 10:33](#event-30637686565)\n\n This was referenced Sep 6, 2026 \n\n[evandhoffman](https://github.com/evandhoffman)added a commit to evanwtf/local-llm that referenced this pull request\n\n[Sep 6, 2026](#ref-commit-cd82bc2)\n\n[ivanfioravanti](https://github.com/ivanfioravanti)marked this pull request as draft\n\n[September 6, 2026 22:22](#event-30655094735)\n\nContributor\nAuthor\n\n| Finalizing Vision, Q2 and tools after rebase. | \n\n[ivanfioravanti](https://github.com/ivanfioravanti)\n\n[force-pushed](https://github.com/antirez/ds4/compare/8f7c4fa65457c1f7d097a8fd5882e8e30c41575e..ffd85d426313ace6dae805e9f2fb4424d5b427fd)the qwen3.8-flash-next branch from\n\n[to](https://github.com/antirez/ds4/commit/8f7c4fa65457c1f7d097a8fd5882e8e30c41575e)\n\n`8f7c4fa`\n\n`ffd85d4`\n[Compare](https://github.com/antirez/ds4/compare/8f7c4fa65457c1f7d097a8fd5882e8e30c41575e..ffd85d426313ace6dae805e9f2fb4424d5b427fd)\n\n[September 6, 2026 23:11](#event-30656157149)\n\n[ivanfioravanti](https://github.com/ivanfioravanti)marked this pull request as ready for review\n\n[September 6, 2026 23:18](#event-30656299619)\n\n[evandhoffman](https://github.com/evandhoffman)added a commit to evanwtf/local-llm that referenced this pull request\n\n[Sep 7, 2026](#ref-commit-4ce9105)\n\n[evandhoffman](https://github.com/evandhoffman)added a commit to evanwtf/local-llm that referenced this pull request\n\n[Sep 8, 2026](#ref-commit-4537b83)\n\n This was referenced Sep 11, 2026 \n\n[ivanfioravanti](https://github.com/ivanfioravanti)\n\n[force-pushed](https://github.com/antirez/ds4/compare/ff4f0ff4fdff70d6b7c3941ef437b91dde960e14..30305544f96cc35357e0dd376eeff2b1ab3e365e)the qwen3.8-flash-next branch from\n\n[to](https://github.com/antirez/ds4/commit/ff4f0ff4fdff70d6b7c3941ef437b91dde960e14)\n\n`ff4f0ff`\n\n`3030554`\n[Compare](https://github.com/antirez/ds4/compare/ff4f0ff4fdff70d6b7c3941ef437b91dde960e14..30305544f96cc35357e0dd376eeff2b1ab3e365e)\n\n[September 12, 2026 23:32](#event-31036732440)\n\n| [@ivanfioravanti](https://github.com/ivanfioravanti) Quick heads-up: the Q4_1 encoder from`a0b1658` appears to have been dropped in`b020c9f` and the squashed`3030554` , while`qwen4_pack.py` still uses it for the PLE sidecar. The current native quantizer marks Q4_1 unsupported. The original implementation likely just needs to be restored before upstreaming. | \n\nContributor\nAuthor\n\n| You rock! Gonna fix this! | \n\n[ivanfioravanti](https://github.com/ivanfioravanti)\n\n[force-pushed](https://github.com/antirez/ds4/compare/d8ba365815af259dc363f991cc1a47dff7637ccb..cf5f223d14ca434142cfbeaaf7eeb9353fee10ba)the qwen3.8-flash-next branch 3 times, most recently from\n\n[to](https://github.com/antirez/ds4/commit/d8ba365815af259dc363f991cc1a47dff7637ccb)\n\n`d8ba365`\n\n`cf5f223`\n[Compare](https://github.com/antirez/ds4/compare/d8ba365815af259dc363f991cc1a47dff7637ccb..cf5f223d14ca434142cfbeaaf7eeb9353fee10ba)\n\n[September 13, 2026 09:21](#event-31048305644)\n\n```\nImplement qwen4exp loading, tokenizer and chat formatting, PLE sidecars,\nMetal text and vision graphs, built-in MTP, and recurrent KV persistence.\nIntegrate the CLI, agent, server, download targets, and conversion tools.\n\nPreserve the native Q4_1 encoder, Q4_K accumulation guard, and shared\nquantizer build target. Keep exact-MTP rewind and vision-cache row-width\nfixes. Persist valid logits and recurrent state at completed prefill chunks,\nand reuse Qwen text-answer history when tools clients omit reasoning.\nRetain exact reasoning replay and test continued checkpoints, cancellation,\ncache eviction, and restart reuse with and without reasoning_content. Include\nreproducible charts and measurements comparing throughput before and after\nthe checkpoint fixes on M3 Ultra.\n\nAdd Qwen trunk steering and prompt activation capture, including GDN layers.\nKeep the MTP predictor outside the 48-row bank and preserve prompt dumps\nthrough one-token prefill tails and multi-row speculative verification.\nRecover Chrome target creation and retain its last page target. Preserve\nfull agent transcripts in stripped-session format when KV no longer fits,\nwhile retaining the newer automatic compaction and continuation logic.\nAdd steering/MTP/capture, Chrome CDP recovery, and full-session save/reload\nregressions. Adapt the intended changes from fork PRs #13 and #7.\n\nReconcile with antirez main bd66c40, preserving DeepSeek v4.1 model IDs,\ntool syntax, session state, and Metal kernel dispatch. Exclude superseded\nengine code and prototypes.\n\nCo-authored-by: Andreas Parusel \nCo-authored-by: david \nCo-authored-by: Michael Kuebbeler \nCo-authored-by: Ogilthorp3 \nCo-authored-by: anemll \n```\n\n[ivanfioravanti](https://github.com/ivanfioravanti)\n\n[force-pushed](https://github.com/antirez/ds4/compare/398920520391d5a8d06c06ee88ad098b669bd76f..ccea7688276a9da8fc1453b13dad3e5c1c044ce2)the qwen3.8-flash-next branch from\n\n[to](https://github.com/antirez/ds4/commit/398920520391d5a8d06c06ee88ad098b669bd76f)\n\n`3989205`\n\n`ccea768`\n[Compare](https://github.com/antirez/ds4/compare/398920520391d5a8d06c06ee88ad098b669bd76f..ccea7688276a9da8fc1453b13dad3e5c1c044ce2)\n\n[September 13, 2026 13:05](#event-31052745223)\n\n This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.\n \n\n[Learn more about bidirectional Unicode characters](https://github.co/hiddenchars)\nAdd this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.", "url": "https://wpnews.pro/news/dwarf-star-support-for-qwen3-8-flash-next", "canonical_source": "https://github.com/antirez/ds4/pull/991", "published_at": "2026-09-16 23:14:39+00:00", "updated_at": "2026-09-16 23:24:03.703782+00:00", "lang": "en", "topics": ["large-language-models", "ai-tools", "developer-tools", "ai-infrastructure"], "entities": ["Qwen3.8 Flash Next", "ivanfioravanti", "antirez/ds4", "evandhoffman", "evanwtf/local-llm", "Hugging Face", "M3 Ultra", "Q4_K imatrix"], "alternates": {"html": "https://wpnews.pro/news/dwarf-star-support-for-qwen3-8-flash-next", "markdown": "https://wpnews.pro/news/dwarf-star-support-for-qwen3-8-flash-next.md", "text": "https://wpnews.pro/news/dwarf-star-support-for-qwen3-8-flash-next.txt", "jsonld": "https://wpnews.pro/news/dwarf-star-support-for-qwen3-8-flash-next.jsonld"}}