Janvitos

Life engineer.

[Janvitos @janvitos]

@BRICSinfo ...and is shot down by Israeli air force.

[Janvitos @janvitos]

Time to buy $ZRO $NEAR $XRP $CRV

[Janvitos @janvitos]

@biancoresearch It seems like $SPCX has already reached the plateau of productivity.

[Janvitos @janvitos]

@BRICSinfo This song started playing in my head when I saw your post https://t.co/yvzTKniJLs

[Janvitos @janvitos]

@WatcherGuru ...or he will threaten to bomb them.

[Janvitos @janvitos]

@KyleHessling1 @AnthropicAI You mean breathtaking like that baby in Seinfeld?

[Janvitos @janvitos]

@decodejar I think the most important part of that capitulation event was the sweep of the low before the subsequent rally.

[Janvitos @janvitos]

@CJ900X Hey @CJ900X, I know ranges are somewhat subjective, but what do you think about this possibility?

[Janvitos @janvitos]

@CJ900X Juicy.

[Janvitos @janvitos]

@VladTheInflator The Divided States of America.

[Janvitos @janvitos]

@JakeKAllDay With ik_llama.cpp, you can get 110 tok/s with an RTX 4070 12GB using Qwen3.6-35B-A3B-IQ4_XS-4.19bpw.gguf (comparable to Q4_K_XL) and MTP. I'm sure any 12GB RTX card can get similar results too.

https://t.co/T4cPZuchEF

[Janvitos @janvitos]

@CryptoCred I would suggest starting with Freqtrade / CCXT for connecting to exchanges and experimenting with trading algos and strategies. Codex (GPT 5.5) is an excellent choice for this. Much more deterministic and accurate than Claude imho.

[Janvitos @janvitos]

@ItsmeAjayKV Curious to know why you chose Q6_K_XL vs Q4_K_XL?

Also, is plan mode an extension?

Thanks!

[Janvitos @janvitos]

@CJ900X Seems like it will close the April 7th gap. Let's see what happens next.

[Janvitos @janvitos]

@decodejar Have you tried Codex (GPT 5.5) with either Codex CLI or Opencode? I've recently switched from Opus/Sonnet, mostly working on trading algos, and it's been day and night for me. Codex is fast, deterministic and doesn't overthink.

[Janvitos @janvitos]

@ItsmeAjayKV For my setup, RTX 4070 Super 12GB, --spec-draft-p-min 0.75 worsens performance, and has no positive impact on draft acceptance.

I'm seeing a clear performance regression in the merged MTP branch. Around 15 tok/sec less and 0.20 acceptance rate less.

[Janvitos @janvitos]

@lyesbahi @danielhanchen @UnslothAI The regression is still present. So I assume this will stay the new normal for the time being.

[Janvitos @janvitos]

@lyesbahi @danielhanchen @UnslothAI MTP has just been merged to the master branch, so I will test and report back.

[Janvitos @janvitos]

@danielhanchen @UnslothAI As I said, I'm pretty sure the issue is related to yesterday's commits that have been applied over the mtp-clean branch.

[Janvitos @janvitos]

@danielhanchen @UnslothAI Tested with --spec-draft-p-min 0.0 and --spec-draft-p-min 0.75 and it has absolutely no impact on acceptance rate or tok/sec.

[Janvitos @janvitos]

@danielhanchen @UnslothAI Just tried with --spec-draft-p-min 0.0 and results are pretty much identical.