正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

MLXs CUDA backend is getting better. It's especially nice if you appreciate fast startup times. But it's also quite fast in general. Here's Qwen3 4B in fp8 running on my DGX Spark. - Processed 18.5k tokens in < 4 seconds - Generates at 32.5 tok/sec with 18.5k context

Awni Hannun

37,721 subscribers

16,542 次观看 • 5 个月前 •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

0 条评论

暂无评论

原始帖子的评论将显示在这里

相关视频

Qwen3-Coder-Flash runs quite fast on an M4 Max with mlx-lm. Running the 4-bit here, generated 4,467 tokens at >107 tokens/sec:

Qwen3-Coder-Flash runs quite fast on an M4 Max with mlx-lm. Running the 4-bit here, generated 4,467 tokens at >107 tokens/sec:

Awni Hannun

196,508 次观看 • 11 个月前

"Why are you benchmarking DGX Spark? It's a training box." Yeah. Low bandwidth, but 128GB of unified memory is just sitting there. Plenty of room to optimize. DGX Spark + Qwen3.6 27B. Four backend/quant combos: 🔴 llama.cpp + UD_Q4_K_XL > 11.0 tok/s (baseline), TTFT 297ms 🟢 llama.cpp + DFlash > 20.4 tok/s (peaks at 97 tok/s), TTFT 320ms 🟡 vLLM FP8 + MTP > 13.1 tok/s, TTFT 540ms 🟣 vLLM NVFP4 + MTP > 24.2 tok/s, TTFT 376ms NVFP4+MTP is the winner for me, rock stable around 24 tok/s, no wild swings. DFlash is the wildcard: massive peaks, but fluctuates a lot. FP8+MTP barely beats baseline, and it's FP8. Love my Spark.

"Why are you benchmarking DGX Spark? It's a training box." Yeah. Low bandwidth, but 128GB of unified memory is just sitting there. Plenty of room to optimize. DGX Spark + Qwen3.6 27B. Four backend/quant combos: 🔴 llama.cpp + UD_Q4_K_XL > 11.0 tok/s (baseline), TTFT 297ms 🟢 llama.cpp + DFlash > 20.4 tok/s (peaks at 97 tok/s), TTFT 320ms 🟡 vLLM FP8 + MTP > 13.1 tok/s, TTFT 540ms 🟣 vLLM NVFP4 + MTP > 24.2 tok/s, TTFT 376ms NVFP4+MTP is the winner for me, rock stable around 24 tok/s, no wild swings. DFlash is the wildcard: massive peaks, but fluctuates a lot. FP8+MTP barely beats baseline, and it's FP8. Love my Spark.

stevibe

39,983 次观看 • 2 个月前

timelapse #85 (27.5 hrs): - currently cant rely on any other coding models except grok code fast 1 + grok 4 fast (for complex reasoning grok 4 fast is 20 cents for 1M tokens) - wrote qwen3-next trainer entirely from scratch to make it more managable - each piece completely done by grok-code-fast-1 in cursor as it seems to handle this task pretty well without the grok 4 fast reasoning - take on smaller problems and complete them quickly (makes it easier with 400 toks/sec over the api) - got distributed fp8 qwen3-next trainer running at 0.8 seconds per step on 8xH100s (still need to finish checkpoint loading logic) - perfect timing as the fp8 version of qwen3-next drops as im writing this - ill be in LA in 2 days (will visit SF mid way through as well) - 12.5% margarita - steak dinner with family - gained intuition on FlashAttention in very long context settings - caught up w/ Kearm h/eng and Arnie Ramesh

timelapse #85 (27.5 hrs): - currently cant rely on any other coding models except grok code fast 1 + grok 4 fast (for complex reasoning grok 4 fast is 20 cents for 1M tokens) - wrote qwen3-next trainer entirely from scratch to make it more managable - each piece completely done by grok-code-fast-1 in cursor as it seems to handle this task pretty well without the grok 4 fast reasoning - take on smaller problems and complete them quickly (makes it easier with 400 toks/sec over the api) - got distributed fp8 qwen3-next trainer running at 0.8 seconds per step on 8xH100s (still need to finish checkpoint loading logic) - perfect timing as the fp8 version of qwen3-next drops as im writing this - ill be in LA in 2 days (will visit SF mid way through as well) - 12.5% margarita - steak dinner with family - gained intuition on FlashAttention in very long context settings - caught up w/ Kearm h/eng and Arnie Ramesh

Elliot Arledge

283,820 次观看 • 9 个月前

DS4 running on DGX Spark (GB10 / CUDA), private branch for now. 12 tokens/sec, the memory bandwidth is limited in this system, at 270GB/sec. But prefill is ways more alighed to M3 Max at ~200 t/s. I'll release when more mature, but it is almost sure that it will get merged.

DS4 running on DGX Spark (GB10 / CUDA), private branch for now. 12 tokens/sec, the memory bandwidth is limited in this system, at 270GB/sec. But prefill is ways more alighed to M3 Max at ~200 t/s. I'll release when more mature, but it is almost sure that it will get merged.

antirez

83,448 次观看 • 2 个月前

First results are in. Llama 4 Maverick 17B active / 400B total is blazing fast with MLX on an M3 Ultra. Here is the 4-bit model generating 1100 tokens at 50 tok/sec:

First results are in. Llama 4 Maverick 17B active / 400B total is blazing fast with MLX on an M3 Ultra. Here is the 4-bit model generating 1100 tokens at 50 tok/sec:

Awni Hannun

149,855 次观看 • 1 年前

Qwen3 235B MoE (22B active) runs so fast on an M2 Ultra with mlx-lm. - 4-bit model uses ~132GB - Generated 580 tokens at ~28 toks/sec

Qwen3 235B MoE (22B active) runs so fast on an M2 Ultra with mlx-lm. - 4-bit model uses ~132GB - Generated 580 tokens at ~28 toks/sec

Awni Hannun

117,763 次观看 • 1 年前

Kimi K2.5 1T runs on 2 M3 Ultras with mlx-lm in it's native precision. It's actually quite usable. Here it's making a space invaders game. Generated 3856 tokens at 21.9 tok/sec using 350GB per machine. Thanks to Tarjei Mandt for the port.

Kimi K2.5 1T runs on 2 M3 Ultras with mlx-lm in it's native precision. It's actually quite usable. Here it's making a space invaders game. Generated 3856 tokens at 21.9 tok/sec using 350GB per machine. Thanks to Tarjei Mandt for the port.

Awni Hannun

241,472 次观看 • 5 个月前

Spent the last few days vibe coding on my NVIDIA DGX Spark. Here's what I learned. Qwen 3.5 122B took one minute and nine seconds to respond "Hi how are you doing". Unusable for vibe coding. Gemma 4 was fast but built a dot instead of a first person shooter game. GPT-OSS 120B was the sweet spot. Fast, capable, and actually produced working HTML. Open source models running locally are not replacing Claude Opus 4.6 or Codex with GPT 5.4. Not even close. But they're getting better every month. The new DGX Spark Bench is live on Real-world benchmarks for local models on local hardware. This is just the start. Full video below.

Spent the last few days vibe coding on my NVIDIA DGX Spark. Here's what I learned. Qwen 3.5 122B took one minute and nine seconds to respond "Hi how are you doing". Unusable for vibe coding. Gemma 4 was fast but built a dot instead of a first person shooter game. GPT-OSS 120B was the sweet spot. Fast, capable, and actually produced working HTML. Open source models running locally are not replacing Claude Opus 4.6 or Codex with GPT 5.4. Not even close. But they're getting better every month. The new DGX Spark Bench is live on Real-world benchmarks for local models on local hardware. This is just the start. Full video below.

BridgeMind

57,065 次观看 • 3 个月前

Running GLM-4.7-Flash with OpenCode locally on M4 Max MacBook Pro. 4-bit model runs at 82 tok/sec. Prefill will get ~4x faster with M5 Max MacBook Pro (~28 Jan). EXO will also support disaggregating prefill and decode across devices, e.g. DGX Spark.

Running GLM-4.7-Flash with OpenCode locally on M4 Max MacBook Pro. 4-bit model runs at 82 tok/sec. Prefill will get ~4x faster with M5 Max MacBook Pro (~28 Jan). EXO will also support disaggregating prefill and decode across devices, e.g. DGX Spark.

Alex Cheema

128,084 次观看 • 5 个月前

GLM-5 runs with mlx-lm on a single 512GB M3 Ultra in Q4. It's quite good in my initial testing and pretty fast as well. It generated a highly functional space invaders game using 7.1k tokens at 15.4 tok/s and 419GB memory. Thanks to Gökdeniz Gülmez and Tarjei Mandt for the port.

GLM-5 runs with mlx-lm on a single 512GB M3 Ultra in Q4. It's quite good in my initial testing and pretty fast as well. It generated a highly functional space invaders game using 7.1k tokens at 15.4 tok/s and 419GB memory. Thanks to Gökdeniz Gülmez and Tarjei Mandt for the port.

Awni Hannun

60,599 次观看 • 4 个月前

Which DGX Spark is running Gemma 4?

Which DGX Spark is running Gemma 4?

Alex Ziskind

25,291 次观看 • 2 个月前

DFlash speculative decoding on Apple Silicon Qwen3.5-9B bf16 · M5 Max · greedy exact match ▸ 85 tok/s, 3.3× at 1024 tokens (runtime) ▸ ~70 tok/s, 2.6× in the video (terminal I/O overhead) ▸ 80 tok/s, 3.1× at 2048 tokens (runtime) Currently working on: → Long context (speedup degrades past 4K tokens, KV cache growth) → Int4 quantized models (27B class) Built on MLX, no CUDA, single machine. Draft generates 16 tokens in parallel, target verifies in one forward pass. Will open source when ready.

DFlash speculative decoding on Apple Silicon Qwen3.5-9B bf16 · M5 Max · greedy exact match ▸ 85 tok/s, 3.3× at 1024 tokens (runtime) ▸ ~70 tok/s, 2.6× in the video (terminal I/O overhead) ▸ 80 tok/s, 3.1× at 2048 tokens (runtime) Currently working on: → Long context (speedup degrades past 4K tokens, KV cache growth) → Int4 quantized models (27B class) Built on MLX, no CUDA, single machine. Draft generates 16 tokens in parallel, target verifies in one forward pass. Will open source when ready.

bstn 👁️

36,942 次观看 • 3 个月前

For the DGX Spark owners. This is what you get with DS4 in your hardware. I want to post this to show how with fast prefill and not very fast generation, the system remains absolutely fine to use.

For the DGX Spark owners. This is what you get with DS4 in your hardware. I want to post this to show how with fast prefill and not very fast generation, the system remains absolutely fine to use.

antirez

39,786 次观看 • 1 个月前

Qwen3.5-35B-A3B running locally on an M4 chip at 49.5 tokens per second. A 35B model. On a laptop. In real time. LOCAL AI IS GETTING SCARY FAST.

Qwen3.5-35B-A3B running locally on an M4 chip at 49.5 tokens per second. A 35B model. On a laptop. In real time. LOCAL AI IS GETTING SCARY FAST.

0xMarioNawfal

478,007 次观看 • 4 个月前

Just for fun, here's what 32 simultaneous long-context generations with Qwen3 Next 80B looks like on an M3 Ultra. Using the new batch generation in mlx-lm. Context size for each is about 5k tokens:

Just for fun, here's what 32 simultaneous long-context generations with Qwen3 Next 80B looks like on an M3 Ultra. Using the new batch generation in mlx-lm. Context size for each is about 5k tokens:

Awni Hannun

50,272 次观看 • 9 个月前

Llama 4 Scout 109B can actually run pretty fast on a 64GB M4 Max using a mixed 3/6 bit quant (4.0 bits-per-weight). Here it is generating 600+ tokens at 42 tok/sec on my laptop!

Llama 4 Scout 109B can actually run pretty fast on a 64GB M4 Max using a mixed 3/6 bit quant (4.0 bits-per-weight). Here it is generating 600+ tokens at 42 tok/sec on my laptop!

Awni Hannun

39,146 次观看 • 1 年前

GLM 4.6 runs quite fast on an M3 Ultra with mlx-lm even at higher precision. Pretty remarkable that it benchmarks competitive to the just-released Sonnet 4.5. Hope those benchmarks hold-up in day-to-day use. Here's a run using 5.5 bpw quantized model, generating 5.3k tokens at 17+ tok/sec using 244 GB. What prompts should I test?

GLM 4.6 runs quite fast on an M3 Ultra with mlx-lm even at higher precision. Pretty remarkable that it benchmarks competitive to the just-released Sonnet 4.5. Hope those benchmarks hold-up in day-to-day use. Here's a run using 5.5 bpw quantized model, generating 5.3k tokens at 17+ tok/sec using 244 GB. What prompts should I test?

Awni Hannun

68,539 次观看 • 9 个月前

Stop doing things that don't make you better. It's you vs your bad habits. Lock in. Sports and entertainment can be good.. but if you aren't happy with where you are at currently it is a distraction.. Put the fast food & the drinks down if you can't look in the mirror with satisfaction.

Stop doing things that don't make you better. It's you vs your bad habits. Lock in. Sports and entertainment can be good.. but if you aren't happy with where you are at currently it is a distraction.. Put the fast food & the drinks down if you can't look in the mirror with satisfaction.

Ashton Hall

635,176 次观看 • 1 年前

What can you do with Llama quality and Groq speed? Instant. That's what. 3 months back: Llama 8B running at 750 Tokens/sec Now: Llama 70B model running at 3,200 Tokens/sec We're still going to get a liiiiiiitle bit faster, but this is our V1 14nm LPU - how fast will V2 be? 😉

What can you do with Llama quality and Groq speed? Instant. That's what. 3 months back: Llama 8B running at 750 Tokens/sec Now: Llama 70B model running at 3,200 Tokens/sec We're still going to get a liiiiiiitle bit faster, but this is our V1 14nm LPU - how fast will V2 be? 😉

Jonathan Ross

179,815 次观看 • 1 年前