Mia

@RobbiewOnline I agree, I use it everyday. I published it yesterday. github.com/MiaAI-Lab/repo…

@MiaAI_lab RepoPrompt looks like a smart way to optimize token usage when working with LLMs, something we dive deep into in the book. The cost of blindly pasting entire files adds up fast (often £1-4 per session), so selective file inclusion is key. Your XML output approach could pair well with local model routing strategies covered in Chapter 2. amazon.com/dp/B0GYDV3FXD If this was helpful, I’d appreciate a repost and a follow. I’ll be sharing more insights from the book, plus what I’ve learned from applying them in the real world, over the coming weeks.

@QuixiAI @deepseek_ai @StepFun_ai If you need vision then you need vision. But if it's not a must I think DS4-Flash is a better choice.

@MiaAI_lab @deepseek_ai Deepseek v4 Flash is text-only, 284B @StepFun_ai Step 3.7 Flash is a Text + Vision model, 198B The vision and the smaller size are more appealing. I choose Step 3.7 Flash.

DeepSeek-v4-Flash beats Step-3.7-Flash in head-to-head tool calling benchmark. Full results in: github.com/MiaAI-Lab/Deep…

@M_Chimiste @QuixiAI @deepseek_ai @StepFun_ai I wish I had the compute to run MiniMax M3. For now, DeepSeek-v4-Flash is unbeatable for 2x DGX Spark setup.

@QuixiAI @MiaAI_lab @deepseek_ai @StepFun_ai I had a lot of tool call issues with Step 3.7. I think I was using Q8 at the time in Hermes Agent. I ended up reverting to Minimax M2.7 and working on moving to M3 for the multimodal input.

I find default DS4-Flash temperature is giving it "too" much creativity for coding. It sometimes does thing I don't ask. Going to run it with temp 0.6 and top_p 0.95 for a while and compare.

@anuntrapid_auto @0xSero Yeah 4 is the way. No going to commit to 3 sparks if I'm not planning to get the 4th one.

@MiaAI_lab @0xSero 4 x is the way. TP 8 vllm/sglang

MiniMax-M3-NVFP4 running on 4x RTX PRO 6000 Repo coming soon.

@theo @WIRED They live in the past.

@WIRED Why are you posting 4 day old articles that are no longer accurate lol

Anthropic is releasing Claude Mythos 5 to trusted organizations and Claude Fable 5 to the public, a version it says can’t be used for cyberattacks. wired.com/story/anthropi…

@mr_r0b0t @Tech2Wild @NVIDIAAI 3 can't be used for TP though... you want it for more concurrent sessions and more kv cache?

@Tech2Wild @NVIDIAAI 3 is a big unlock tho ngl! Bet you're already thinking about 4 and a switch tho 😛🤩 Imma try to surprise those urges for a bit

So I did a thing 😁

@joaosump @0xSero Yes, TP works in pairs, 2,4, etc. 3 units would be good more concurrencies and kv-cache.

@MiaAI_lab @0xSero Damn, that's nice to know, have been considering 2x and thought that was the ceiling. 3x would be an odd number though for vllm right? :(

@joaosump @0xSero No, you go up to 3x DGX Sparks connected directly without the need for a switch. There's even a workaround to connect 4x DGX Sparks without a switch.

@MiaAI_lab @0xSero Btw, wouldn't 3x sparks require an expensive switch?

@0xSero Any solution to connect to Codex app from my iPhone when using codex-shim?

@advented_ Yes, but it works remarkably well. I tried with 4 sessions, and max tok/s was around 80, which is insane considering the limited bandwidth of DGX Spark.

@MiaAI_lab They way I understand is Sequences mean concurrent requests. So your recipe can serve 6 requests at 1M context concurrently. Under full/heavy load t/s can spike and drop bc the longest 1M session has to finish generating for other concurrent requests to continue

53 tok/s achieved on Step-3.7-Flash NVFP4 with MTP on 2x DGX Spark with 256k context. 🎉🥳 Elapsed time: 56.694 s Prompt tokens: 29 Generated tokens: 3000 Total tokens: 3029 Generation tok/sec: 52.92 End-to-end tok/sec: 53.43

@AlicanKiraz0 Is this with MTP?

nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4 and 4x DGX Spark 430 GB Vram 🔥 25 tok/sec 🚀

@TheAhmadOsman That's a lot of GPUs. Looks messy as hell.

My house has 33 GPUs. > 21x RTX 3090s > 4x RTX 4090s > 4x RTX 5090s > 4x Tenstorrent Blackhole p150a Before AGI arrives: Acquire GPUs. Go into debt if you must. But whatever you do, secure the GPUs.

@usr_bin_roygbiv @tmaiaroto Not to mention the amount of power and heat that would come out of such a config.

@usr_bin_roygbiv @tmaiaroto Do tell how many 5090 or 3090 I would need to run DeepSeek-v4-Flash with 1M context. Meanwhile, you can do that with 2x DGX Sparks with 45 tok/s. x.com/MiaAI_lab/stat…

Do NOT buy a DGX Spark if prefill speeds and running local AI is not important to you.

@BlockedPaths I love it and with 1M context, locally, it's a no brainer IMO.

@MiaAI_lab Deepseek is so slept on.

@DevRico003 @TheGoldenAnvil I'm doing coding with it almost all day long. It's awesome.

@MiaAI_lab @TheGoldenAnvil can you do some real coding work with DS4F on 2x sparks? I'm considering to buy a second spark but I'm not sure yet.

Awaiting update, seed is growing the the interim.

@broadfield_dev It's awesome, it's my go-to right now. And 1M context it can do long /goals with Codex app.

@MiaAI_lab deepseek v4 flash is great, I use it a lot in my DIY agent harness and there is nothing it can't handle given enough time. It's also not painful to give it all the time it needs to complete tasks.

@ljupc0 Can't really compare bcs Step's max context is 256k vs 1M for the DS4-Flash

@MiaAI_lab How does the Step model fare with increasing context depth? I've tested DeepSeek and it's excellent. It takes context 784K to drop the TG to <10 tok/s (on M2 Max 96gb) #context-depth-benchmarks-via-llama-benchy" target="_blank" rel="nofollow noopener">github.com/ljubomirj/ds4/…

Local agentic 'Tool-Call Benchmark' between DeepSeek-v4-Flash to Step-3.7-Flash. Same host, same 69 scenarios, two models. Results: DeepSeek-v4-Flash: 90/100 quality, 59 passed, 6 partial, 4 failed Step-3.7-Flash: 87/100 quality, 55 passed, 10 partial, 4 failed 👇