{"slug": "dflash", "name": "DFlash", "category": "ai", "category_name": "AI tools and apps", "description": "", "recommenders": 3, "commenters": 5, "warners": 0, "score": 3.15, "rising": 0.35, "first": "2026-04-14", "latest": "2026-09-21", "why": "mtp or dflash help a lot and should be used when checking what is useful", "why_is_quote": true, "has_verdict": false, "sentiment": {"recommend": 3, "neutral": 0, "compare": 2, "warn": 0}, "spark": [0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 1, 0], "listed": true, "loved": [], "caveat": "", "short_quote": "mtp or dflash help a lot and should be used when checking what is useful", "kind": "", "icon": false, "domain": "z-lab.ai", "github_repo": null, "official_url": "", "pricing": "", "aliases": ["DFlash"], "verdict": null, "praise": [{"label": "", "summary": "", "quotes": [{"id": 796645344, "item_id": 49790334, "url": "https://news.ycombinator.com/item?id=49790334", "by": "alex7o", "date": "2026-09-21", "version": null, "stance": "recommend", "quote": "mtp or dflash help a lot and should be used when checking what is useful", "excerpt": "On my m5 max 27b model does 75tps on 256k ctx and starts at 80 on the 8k ctx when you add https://huggingface.co/collections/z-lab/dflash-2 to it. So yeah base might be 30tps (I used iq4) but mtp or dflash help a lot and should be used when checking what is useful and what is not for running models as it is not fare to judge without them.", "context": "M5 Max user", "use_case": "speculative decoding speedup", "switched_from": "", "story": "M5 Ultra Mac Studio Review"}, {"id": 781858227, "item_id": 48866139, "url": "https://news.ycombinator.com/item?id=48866139", "by": "verdverm", "date": "2026-07-10", "version": null, "stance": "recommend", "quote": "Z Lab had been publishing dflash addons and boosting their tgen 2-3x.", "excerpt": "Second this, they keep getting performance upgrades too. Z Lab had been publishing dflash addons and boosting their tgen 2-3x. I'm looking at doing comparative evals right now https://huggingface.co/collections/z-lab/dflash", "context": "", "use_case": "", "switched_from": "", "story": "Don't discontinue Gemini 2.5 Flash"}, {"id": 781782096, "item_id": 48861381, "url": "https://news.ycombinator.com/item?id=48861381", "by": "verdverm", "date": "2026-07-10", "version": null, "stance": "recommend", "quote": "DFlash addon improves tgen speed 2-3x.", "excerpt": "Unconvincing, I use this model for agentic work, and while it's obviously not as good, this person is using their personal setup and experience as a justification. 1. Mac is slow at prompt processing, the Spark is much better for coding. Most of the time in coding is spent in prompt processing. 2. DFlash addon improves tgen speed 2-3x. (https://huggingface.co/collections/z-lab/dflash) 3. Harness matters, they gave no", "context": "uses Qwen 27B for agentic work", "use_case": "speeding up local LLM generation", "switched_from": "", "story": "No, Qwen 27B isn't the sweet spot"}, {"id": 781592400, "item_id": 48849525, "url": "https://news.ycombinator.com/item?id=48849525", "by": "verdverm", "date": "2026-07-09", "version": null, "stance": "recommend", "quote": "tgen is way up", "excerpt": "z-lab has been dropping dflash addons for a lot of models https://huggingface.co/collections/z-lab/dflash I'm running the qwen3.6-27B + dflash on a spark and tgen is way up, but keep the number low, acceptance rate is terrible beyond half a dozen and it requires more memory", "context": "", "use_case": "Speculative decoding add-on", "switched_from": "", "story": "Hy3"}, {"id": 767229889, "item_id": 47951868, "url": "https://news.ycombinator.com/item?id=47951868", "by": "simjnd", "date": "2026-04-29", "version": null, "stance": "recommend", "quote": "can provide an uplift up to 5x in decoding", "excerpt": "> The one thing I would want everyone curious about local LLMs to know is that being able to run a model and being able to run a model fast are two very different thresholds. You can get these models to run on a 128GB Mac, but we need to first tell if Q4 retains enough quality (models have different sensitivities to quantization) and how fast it runs. Very valid. This is an active area of research, and there are a lo", "context": "", "use_case": "speculative decoding", "switched_from": "", "story": "Mistral Medium 3.5"}]}], "caveats": [], "who": [{"by": "alex7o", "context": "M5 Max user", "url": "https://news.ycombinator.com/item?id=49790334", "date": "2026-09-21"}, {"by": "verdverm", "context": "uses Qwen 27B for agentic work", "url": "https://news.ycombinator.com/item?id=48861381", "date": "2026-07-10"}], "tags": [], "use_cases": ["speculative decoding speedup", "speeding up local llm generation", "speculative decoding add-on", "speculative decoding"], "compared": [{"slug": "qwen", "name": "Qwen", "co": 3, "switched_from": 0, "switched_to": 0}], "makers": [], "threads": [{"url": "https://news.ycombinator.com/item?id=49787313", "title": "M5 Ultra Mac Studio Review", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48864507", "title": "Don't discontinue Gemini 2.5 Flash", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48860979", "title": "No, Qwen 27B isn't the sweet spot", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48847552", "title": "Hy3", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48024540", "title": "Accelerating Gemma 4: faster inference with multi-token prediction drafters", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=47949642", "title": "Mistral Medium 3.5", "mentions": 1}], "sentiment_by_month": {"2025-11": {}, "2025-12": {}, "2026-01": {}, "2026-02": {}, "2026-03": {}, "2026-04": {"compare": 1, "recommend": 1}, "2026-05": {"compare": 1}, "2026-06": {}, "2026-07": {"recommend": 1}, "2026-08": {}, "2026-09": {"recommend": 1}, "2026-10": {}}, "versions": [], "version_unnamed": 0, "family_ref": null, "runs_on": null, "lines": [], "quotes": [{"id": 796645344, "item_id": 49790334, "url": "https://news.ycombinator.com/item?id=49790334", "by": "alex7o", "date": "2026-09-21", "version": null, "stance": "recommend", "quote": "mtp or dflash help a lot and should be used when checking what is useful", "excerpt": "On my m5 max 27b model does 75tps on 256k ctx and starts at 80 on the 8k ctx when you add https://huggingface.co/collections/z-lab/dflash-2 to it. So yeah base might be 30tps (I used iq4) but mtp or dflash help a lot and should be used when checking what is useful and what is not for running models as it is not fare to judge without them.", "context": "M5 Max user", "use_case": "speculative decoding speedup", "switched_from": "", "story": "M5 Ultra Mac Studio Review"}, {"id": 781858227, "item_id": 48866139, "url": "https://news.ycombinator.com/item?id=48866139", "by": "verdverm", "date": "2026-07-10", "version": null, "stance": "recommend", "quote": "Z Lab had been publishing dflash addons and boosting their tgen 2-3x.", "excerpt": "Second this, they keep getting performance upgrades too. Z Lab had been publishing dflash addons and boosting their tgen 2-3x. I'm looking at doing comparative evals right now https://huggingface.co/collections/z-lab/dflash", "context": "", "use_case": "", "switched_from": "", "story": "Don't discontinue Gemini 2.5 Flash"}, {"id": 781782096, "item_id": 48861381, "url": "https://news.ycombinator.com/item?id=48861381", "by": "verdverm", "date": "2026-07-10", "version": null, "stance": "recommend", "quote": "DFlash addon improves tgen speed 2-3x.", "excerpt": "Unconvincing, I use this model for agentic work, and while it's obviously not as good, this person is using their personal setup and experience as a justification. 1. Mac is slow at prompt processing, the Spark is much better for coding. Most of the time in coding is spent in prompt processing. 2. DFlash addon improves tgen speed 2-3x. (https://huggingface.co/collections/z-lab/dflash) 3. Harness matters, they gave no", "context": "uses Qwen 27B for agentic work", "use_case": "speeding up local LLM generation", "switched_from": "", "story": "No, Qwen 27B isn't the sweet spot"}, {"id": 781592400, "item_id": 48849525, "url": "https://news.ycombinator.com/item?id=48849525", "by": "verdverm", "date": "2026-07-09", "version": null, "stance": "recommend", "quote": "tgen is way up", "excerpt": "z-lab has been dropping dflash addons for a lot of models https://huggingface.co/collections/z-lab/dflash I'm running the qwen3.6-27B + dflash on a spark and tgen is way up, but keep the number low, acceptance rate is terrible beyond half a dozen and it requires more memory", "context": "", "use_case": "Speculative decoding add-on", "switched_from": "", "story": "Hy3"}, {"id": 768478336, "item_id": 48029896, "url": "https://news.ycombinator.com/item?id=48029896", "by": "egeres", "date": "2026-05-05", "version": null, "stance": "compare", "quote": "I can't wait to see how it will compare against MTP", "excerpt": "There's also growing interest on integrating DFlash: https://github.com/ggml-org/llama.cpp/issues/21978, I can't wait to see how it will compare against MTP", "context": "", "use_case": "", "switched_from": "", "story": "Accelerating Gemma 4: faster inference with multi-token prediction drafters"}, {"id": 767229889, "item_id": 47951868, "url": "https://news.ycombinator.com/item?id=47951868", "by": "simjnd", "date": "2026-04-29", "version": null, "stance": "recommend", "quote": "can provide an uplift up to 5x in decoding", "excerpt": "> The one thing I would want everyone curious about local LLMs to know is that being able to run a model and being able to run a model fast are two very different thresholds. You can get these models to run on a 128GB Mac, but we need to first tell if Q4 retains enough quality (models have different sensitivities to quantization) and how fast it runs. Very valid. This is an active area of research, and there are a lo", "context": "", "use_case": "speculative decoding", "switched_from": "", "story": "Mistral Medium 3.5"}, {"id": 764387024, "item_id": 47774189, "url": "https://news.ycombinator.com/item?id=47774189", "by": "xiphias2", "date": "2026-04-14", "version": null, "stance": "compare", "quote": "How does this compare to DFlash?", "excerpt": "How does this compare to DFlash? https://z-lab.ai/projects/dflash/ And DDTree? https://liranringel.github.io/ddtree/", "context": "", "use_case": "", "switched_from": "", "story": "Introspective Diffusion Language Models"}]}