Ankit Prateek

Reader. Geek. Gamer.

[ᴅᴀɴɪᴇʟ ᴍɪᴇssʟᴇʀ 🛡️ @DanielMiessler]

Share your inner interests. Like a child.

(Only don’t lose it when you grow up)

And if you already lost it, dig deep and find it again.

“The most personal is the most creative.” Martin Scorsese

[Pavlo Molchanov @PavloMolchanov]

What if you could take three completely different model families… and distill them into one tiny model? 🤯

📜 Paper: https://t.co/K2iKD4xFvp

MOPD (Multi-Teacher On-Policy Distillation) has become a standard procedure in post-training. We already distill multiple specialized variants of the same model into a single set of weights.
But what if we could go further - and distill models from entirely different families? Turns out, it is possible.

Today we’re releasing a paper on cross-tokenizer distillation - our first steps in this exciting direction. 📄

We distilled Qwen3-4B, Phi-4-Mini, and Llama-3B into Llama-3.2-1B.
MMLU jumped from 32.05 → 46.32 when using multiple teachers. 📈

The team is now working on Nemo-RL integration so the community can try this method in their own settings. Plus, we are scaling experiments up. 🚀

[Ankit Prateek @ankit_appy]

@opencode MiMo-2.5

[Ankit Prateek @ankit_appy]

@MichaelHyatt @Teknium @Hermes_agentAI @gregisenberg @AlexFinn @NetworkChuck Install on your main machine for daily use. It gets better the more you use it.
Also install on a machine that you'll keep running 24x7, for your cronjobs, and for using it remotely via telegram, discord, WhatsApp. Hermes is probably the best harness out there.

[Ankit Prateek @ankit_appy]

Complete global data-flow tracking down to external method bodies, zero brute-force prompt stuffing, and a cloud bill ~$2.
Smart scaffolding > massive compute budgets.

Lesson learned: Don't scan Chromium again. It's too good. 😂

[Ankit Prateek @ankit_appy]

~21h grinding a SAST run on Chromium's net/ alongside its full dependency graph.
283M total tokens (~163M DeepSeek + ~120M local LLM). The structural orchestration framework built for deep context caching hit a 97.8% cloud cache-hit rate (only ~3.5M cache misses).

[Ankit Prateek @ankit_appy]

Called it a discount. Turns out it was the price.
#DeepSeek just made the 75% V4-Pro cut permanent.
The token economics I wrote about 5 days ago aren't a promo window anymore – they're the floor. Resharing: https://t.co/ozuUytMPSI

[Ankit Prateek @ankit_appy]

33 Million Tokens for $0.25

Just ran a full SAST scan against 1M+ lines of code for the price of a gumball.
The secret? Hybrid Architecture + Context Caching.
Master: DeepSeek V4-Flash (Orchestrator) Worker: llama.cpp (Local)
Full deep dive on LinkedIn: [https://t.co/nQzJq0THP7]

[Ankit Prateek @ankit_appy]

If your framework supports MTP, turn it on. It’s an uncompromised velocity multiplier for repository-wide code reviews and massive doc analysis.

Full writeup: https://t.co/1Tx2M4DnMN

Shoutout to @ggerganov & team for https://t.co/kvGqXzAlJD

[Ankit Prateek @ankit_appy]

Multi-Token Prediction (MTP) is a rare "free lunch" in LLM inference.

Just finished benchmarking Qwen 3.6 27B on a single RTX 5090 using llama.cpp. At extreme context scales, MTP roughly DOUBLES generation throughput with 0 quality loss.

The raw telemetry 👇 🧵

[Ankit Prateek @ankit_appy]

Is it lossy? No. At temp=0, MTP holds a strict veto layer. The core model brain validates every parallel guess before it hits your screen.

You get cosmetic word changes from FP near-ties, but logical, code, & math accuracy remain 100% intact.

[Ankit Prateek @ankit_appy]

@r0ktech what are you talking about? second one is pure gold https://t.co/EB9zfCNcUs 😂

[Ankit Prateek @ankit_appy]

@KaiXCreator Ignorance. Outrage.

[Ankit Prateek @ankit_appy]

@neural_avb Thanks for sharing the insights.
I gave up on opencode a while back after it's timeout failed repeatedly while working with llama.cpp

[Ankit Prateek @ankit_appy]

33 Million Tokens for $0.25

Just ran a full SAST scan against 1M+ lines of code for the price of a gumball.
The secret? Hybrid Architecture + Context Caching.
Master: DeepSeek V4-Flash (Orchestrator) Worker: llama.cpp (Local)
Full deep dive on LinkedIn: [https://t.co/nQzJq0THP7]

[The Shift Journal @TheShiftJournal]

5 Signs of Genuinely Good Person...

[ClaudeDevs @ClaudeDevs]

Claude Code weekly limits are increasing 50%, now through July 13.

Live now for all Pro, Max, Team, and seat-based Enterprise users.

[M @mxn7q]

#GameOfThrones

برد خاطري اخيرا