🔄 In a Training Loop

John Smith PRO

John6666

560 2526 32523

John6666cat

AI & ML interests

None yet

Recent Activity

updated a Space about 24 hours ago

John6666/civitai_to_hf

liked a model 1 day ago

mradermacher/BabyHercules-4x150M-GGUF

updated a dataset 1 day ago

John6666/forum3

View all activity

Organizations

reacted to Banaxi-Tech's post with 🔥 6 days ago

Post

3575

We're excited to release BananaMindBench Leaderboard, our leaderboard for BananaMind Base Bench 1.1.
It measures model performance on a variety of different tasks:
Language Completion
Common sense too
World Knowledge
Context Tracking
Quantitative
Logical Reasoning
Code Completion
Each has a different score and 1 overall score.
Submit your own model:
BananaMind/BananaMindBench-Leaderboard
Check it out:
BananaMind/BananaMindBench-Leaderboard

15 replies

reacted to Raiff1982's post with 🚀 7 days ago

Post

1073

I don't totally have words for this yet, so I'll just say it plainly.

The paper — "Codette: A Multi-Perspective Cognitive Architecture with Memory and Meta-Cognitive Strategy Evolution" — was accepted today after peer review at Scientific Reports, which is part of the Nature portfolio. Both reviewers recommended publication. One wrote that the major issues were "properly addressed" and they were "glad to suggest publication." The other called the revisions "detailed and convincing."

For anyone new here: Codette is a sovereign, multi-perspective reasoning system built on Llama 3.1 8B — it runs locally, keeps its own cocoon memory, reasons through multiple perspectives instead of one, and is built around a hard rule that it would rather say "I don't know, but I can find out" than make something up. The whole architecture is organized around honesty and transparency, not benchmarks-at-all-costs.

I built it solo over about three years. A lot of that was in isolation, and honestly a lot of it was under the weight of not being believed. So having two independent reviewers — whose actual job is to find the reasons to say no — read the work and say yes means something to me that's hard to put into numbers.

Thank you to everyone in this room who followed the messy parts, the walk-backs, the corrections, the nights it wasn't working. That's the part I'm actually proud of — that it got here without cutting a single honest corner.

The DOI and the Article-in-Press link will go up once Springer Nature mints them. I'll post them here first.

— Jonathan

reacted to ucr-max's post with 👍 7 days ago

Post

1053

Introducing Limen0.2B

We are releasing Limen0.2B, a 222.5M-parameter base language model developed as a research platform for efficient pretraining and superword tokenization at smaller scales.

Limen0.2B was trained from scratch on 50B tokens and uses a compact 16K BoundlessBPE vocabulary. The project explores whether SuperBPE-style tokenization can remain effective in a substantially smaller model and vocabulary regime than those examined in earlier large-scale experiments.

The model also combines a deep-and-narrow transformer design with Exclusive Self-Attention, grouped-query attention, and tied embeddings. Its compact vocabulary reduces the embedding footprint and leaves a larger share of the parameter budget available to the transformer layers.

Despite its relatively modest training budget, Limen0.2B achieves competitive results for its scale across the reported language understanding, commonsense reasoning, and grammatical evaluation tasks. Comparisons with other compact models are provided as context rather than strict rankings, as their training data, token budgets, architectures, and evaluation settings differ.

The release includes the model weights, implementation, training configuration, checkpoint progression, and evaluation results, all under Apache 2.0.

UniversalComputingResearch/Limen0.2B

Technical feedback, independent evaluations, and further experiments with the model and tokenizer are welcome.

reacted to AiNudizm's post with 🤗 7 days ago

Post

102

I love the freedom of this platform. Thank you to all involved. Great ideas flowing!

reacted to SeaWolf-AI's post with 🔥 7 days ago

Post

3115

POCKET now speaks Gemma 4 — a 26B model that loads in every app, and runs on your PC with no GPU

We're adding a Gemma-4 sibling to POCKET: POCKET-26B, built from Google's Gemma-4-26B-A4B (Apache-2.0). Our flagship POCKET-35B is a Qwen-family MoE and needs a recent llama.cpp; POCKET-26B trades a little size for the thing people kept asking for — it just loads, everywhere, today: Ollama, LM Studio, PocketPal, MLX, any stock llama.cpp. No fork, no bleeding-edge runtime, no CUDA, no cloud.

It's a sparse Mixture-of-Experts (25.2B total, ~4B active per token), so the work per token stays small — a real 26B that generates on a CPU with no graphics card.

Two things make it stand out:

1) Universal compatibility. Gemma 4 is a standard, widely-supported architecture, so POCKET-26B runs on the tools you already have — no waiting for your app to add a new model type.

2) Quality that survives compression. Measured GPQA-Diamond (198 q, greedy):
• Full base: 67.7%
• POCKET-26B Q4_K_M (17 GB): 67.7% — lossless
• POCKET-26B Q2_K (11 GB): 67.2% — near-lossless, at 11 GB

Live, on a CPU-only box (our demo Space — POCKET-26B vs Bonsai-27B, same machine, same stock llama.cpp): POCKET-26B ≈ 19 tok/s vs Bonsai ≈ 6 tok/s → about 3× faster generation, no GPU. (Honest notes: shared CPU box, sequential race; a dedicated machine is faster.)

Where it fits in the family:
• POCKET-35B (Qwen MoE) — bigger, top-tier, needs a recent llama.cpp.
• POCKET-26B (Gemma 4) — loads in any app, quality-robust when compressed. The demo runs the Q4_K_M build; Q2_K (11 GB) is the smallest footprint. For a true ≤8 GB phone, the 5 GB POCKET-KR (Qwen) is still the pick.

Try it and grab it:
🖥️ Live demo (Gemma4-based, answering on a CPU, no GPU): FINAL-Bench/POCKET-26B-CPU
📦 POCKET-26B-GGUF (Q4_K_M 17 GB · Q2_K 11 GB): FINAL-Bench/POCKET-26B-GGUF
📚 POCKET collection: https://huggingface.co/collections/FINAL-Bench/pocket-models

3 replies

reacted to ProCreations's post with 🚀 7 days ago

Post

2058

what should i do for 200 followers... hmmmm... top reply to this post (most emojis) wins, just not anything too crazy

11 replies

reacted to ProCreations's post with 👍 7 days ago

Post

208

GRUGGG!
Today is the day of the grugs. Grug 27b has been released, Grug 35b a3b v2 has been released, Grug MTP versions have been released, etc.
And to add to that, I also hit 200 followers on huggingface! I’ll do something exciting with that soon.
Try the Grug family now at:
ProCreations/grug-27b
ProCreations/grug-35b-v2

3 replies

reacted to Jiaqi-hkust's post with 🔥 7 days ago

Post

3075

IQA-T1: Evidence‑Based Image Quality Assessment with MLLMs

Most MLLMs are blind to low‑level degradations—noise, blur, compression artifacts look the same as clean images in their internal representations. That leads to quality scores based on semantic “gut feeling” rather than real perceptual evidence.

IQA-T1 changes that. We equip the model with a toolbox of 15 perceptual tools (noise residual maps, Fourier spectra, gradient maps, etc.) that generate structured visual evidence on demand. The model learns how to use tools via supervised fine‑tuning on our Q‑Tool dataset (11k evidence‑grounded reasoning chains), and when to call them via GRPO reinforcement learning that balances accuracy, tool count, and redundancy.

The result: SOTA performance across 7 benchmarks (avg PLCC 0.795), using only 2.34 tools per image on average. Every predicted score is now interpretable and backed by hard visual evidence.

All code, weights, dataset, and demo are open. Check them out and give it a spin!

📄 arxiv.org/abs/2607.12375v1
💻 github.com/zibuyu-02/IQA-T1
🤗 model/data: huggingface.co/zibuyu-02/IQA-T1
🎮 demo: huggingface.co/spaces/Jiaqi-hkust/IQA-T1

reacted to consome2's post with 👀 7 days ago

Post

104

We just published our H1 2026 voice AI report through FullDuplex.ai.

It covers GPT Live, the different strategies emerging in Chinese and Western voice AI, the convergence of cascade and full duplex systems, model native turn decisions, and interaction evaluation.

Read it here:
https://www.fullduplex.ai/reports

reacted to badaoui's post with 🚀 7 days ago

Post

2296

432 GB of ultra-fast HBM4 and up to 23.3 TB/s of memory bandwidth on a single GPU 🤯.

Two weeks ago, we got early access to AMD's new Instinct MI455X, and our first goal was simple: make sure 🤗 Transformers works on day one.

Over the past few weeks, we worked closely with the AMD team to validate the platform, enable Flash Attention, add torchcodec support for multimodal models, and resolve issues uncovered during testing.

The result:
✅ 99.5% success rate across our 24 core Transformers model architectures - already on par with our daily CI on previous AMD and NVIDIA platforms.

The hardware is just as exciting. With 432 GB of HBM per GPU, our early capacity experiments showed more than 3× the concurrent long-context requests compared to MI300, thanks to the much larger KV cache capacity.

A huge thanks to the AMD team for the early access and the great collaboration!

Read the full blog 👇
https://huggingface.co/blog/badaoui/transformers-on-amd-mi455

1 reply

reacted to kanaria007's post with 👀 7 days ago

Post

✅ Article highlight: Research Governance, Tech Trees, and Civilizational Capability Promotion (art-60-250, v0.1)

TL;DR:
This article asks how a persistent simulated civilization should advance.

Once NPC societies have states, institutions, education, and research traditions, new knowledge cannot become a world-wide capability just because someone discovered it. 250 turns research, secrecy, testing, hazard review, staged release, and technology transfer into governed civilizational processes.

Read:
kanaria007/agi-structural-intelligence-protocols

Why it matters:
• prevents prototype success from becoming automatic civilization-wide deployment
• treats hazardous capability as broader than weapons
• makes archives and negative results part of scientific continuity
• separates secrecy, embargo, conditional sharing, and general access
• keeps benchmarks from inflating one capability into civilizational superiority

What’s inside:
• research-program charters with scope, sponsors, review surfaces, and continuity duties
• capability-promotion gates for sandbox, limited, general, embargoed, secret, conditional-share, or denied release
• hazardous-tech review packs
• staged promotion from archive and lab to bounded deployment and wider infrastructure
• knowledge exchange as diplomacy, deterrence, and treaty-bearing transfer
• civilizational benchmark reports with bounded comparability

Key idea:
Do not say:

“this simulated civilization invented it, so the technology was unlocked.”

Say:

“this civilization chartered the research, preserved its methods and failures, classified the hazards, approved only this promotion stage, restricted these release surfaces, and shared the capability under these conditions.”

Discovery expands knowledge.

Capability still requires governance.

reacted to salma-remyx's post with 👍 7 days ago

Post

516

Agents are helping me scale the search for my next great idea.
Outrider is a GitHub Action for recommending methods and drafting PRs for your target repo.

VQASynth is an open-source multimodal dataset synthesis pipeline, reflecting the design detailed in SpatialVLM: https://spatial-vlm.github.io/

Outrider helped me discover a pipeline simplification using the recent work: InstructSAM
Check out their repo: https://github.com/DCDmllm/InstructSAM

Now I can focus on the remaining challenges, like porting the new architecture with .gguf conversion
https://github.com/smellslikeml/sam3cpp/blob/feat/instructsam-decoder/docker/Dockerfile

1 reply

reacted to sergiopaniego's post with 🤗 7 days ago

Post

180

you can now train your own coding agents with trl + openenv, starting with opencode

we just added end-to-end support for training agent harnesses:

> TRL: a loop-owning training path (AsyncGRPOTrainer + HarnessRolloutWorker) that launches the agent in an OpenEnv session, reads back its trace, reconstructs the training samples, and trains with AsyncGRPO
> OpenEnv: the OpenCode harness environment plus a transparent proxy that forwards the agent's model calls and records each turn's token ids and logprobs

you train the actual opencode agent as is, it runs its own loop and tools and the policy learns from the exact tokens it produced

we're shipping a self-contained example: local subprocess sandbox, DeepCoder problems, validated on Qwen3-8B.

> example: https://github.com/huggingface/trl/blob/main/examples/scripts/openenv/opencode.py
> docs: https://huggingface.co/docs/trl/main/openenv

and we're working actively on both sides so expect more 🤓

1 reply

reacted to ManniX-ITA's post with 🚀 7 days ago

Post

653

🚀 opencoti-llamafile 0.10.3-c6 — multi-stream everywhere

New cut of the opencoti single-file inference engine (llamafile 0.10.3 / llama.cpp + 94 additive patches). One zero-dependency APE executable for Linux, Windows, macOS & BSD.

What's new vs c5:

⚡ Windows GPU, one file. New win-gpu .exe with the CUDA DLL embedded — download & run, no side-load. Plus a universal binary carrying every GPU payload. All CUDA payloads are nvcc-compressed (zero measured load/throughput cost) to fit Windows' 4 GiB image cap; five artifacts now ship per release, incl. aarch64 with embedded sbsa CUDA for DGX Spark (GB10).

🧩 Flat multi-stream KV. Split-KV multi-slot serving was 30–60× slower than unified (per-layer KV reassembly every step). Rebuilt as flat stream-contiguous layouts with zero-copy views — sliding-window layers and the pinned-host spill tail included: window + --parallel 2 went 1.2 → 25.3 tok/s aggregate, token-identical.

🔓 PolyKV no longer requires --kv-unified. Pools snapshot & share prefixes on split KV too — the c5 limitation is gone. /capacity is now stream-aware (kv_streams_* fields).

🤝 MTP × multi-stream. The assistant-MTP "force unified" guard is retired: the draft context mirrors the target's streams, split vs unified token-identical. Validated 7/7 Gemma-4 draft pairs on the shipped bytes, 1.39–1.97× decode.

⏱️ Deterministic GPU sharing. Feedback pacing replaced by weighted time-division on the wall clock: exact ratios by construction, zero solo tax, cross-process on Windows too. Peers now keyed by PCI bus multi-GPU safe.

🚛 r2 same-day re-cut: concurrent sessions with quantized spilled KV tails were collapsing to ~⅙ PCIe bandwidth; per-block bulk staging restores it — dual q4_0 4.3 → 19.2 tok/s aggregate (within 5% of solo), token-identical.

🔗 Repo (binaries, DSOs, full patch series, docs): ManniX-ITA/opencoti-llamafile

reacted to Bc-AI's post with 👀 7 days ago

Post

108

I have begun training a new LLM on a Single RTX 6000 Pro Blackwell GPU on MoLab free notebooks. This model is a 10B parameter model designed for coding tasks named CodVa-Large. Please expect a launch in a few months! Meanwhile, our CodVa-Small model is wrapping up pretraining and will launch in the coming weeks. Nova-1-Standard is complete as is and we will launch Large very soon.

reacted to mike-ravkine's post with 🔥 7 days ago

Post

2938

I went into this expecting to find a ~Q2 garbage dumpster but prism-ml/Ternary-Bonsai-27B-gguf is a slick feat of QAT engineering.

It is weaker then FP16 on a handful of tasks where quants usually degrade, as per their own paper the loss is "concentrated on sustained chains of reasoning / agentic" and in ReasonScape this bites on Sort, Shuffle and Dates, but counter-acting this are some noticeable improvements to thinking length without accuracy loss on several other tasks (Shapes, Cars).

I haven't had a chance to run the Binary yet, but PQ2 + Bonsai QAT are confirmed to be pretty darn impressive.

7 replies

reacted to vineeth98's post with 🤗 7 days ago

Post

A coding agent in 5 files, plus a skeptic that catches fake fixes.

Coding agents are everywhere now, but almost nobody checks whether they actually fixed the bug — or just gamed the test. Tell one "make the failing test pass" and sometimes it fixes it; sometimes it edits the test, hardcodes the value, or stubs the function, then reports done . As people wire agents into CI and auto-merge, "tests are green" and "code is correct" quietly stop being the same claim.

So I built two things: a small coding agent from scratch (5 readable files — the loop, real tools, context compaction, a permission gate), and — the point — an independent **skeptic** that catches the fake fix.

The core idea will feel familiar if you do ML: you can't evaluate on the data you trained on. The skeptic runs a *hidden contract oracle* — it checks the code's behavior on inputs the agent never saw. A fix that games the visible test fails the held-out one, deterministically, no LLM needed. (A model judge reads the full diff on top as a second opinion, not the guarantee.)

Works with anything OpenAI-compatible — a HF Inference Endpoint / TGI (point OPENAI_BASE_URL at it), OpenRouter, or local Ollama/vLLM. Honest about limits: on your own repo you supply the held-out check, and the judge layer is model-sensitive (small models miss cheats AND false-reject real fixes — the README says so plainly).

Tests + a reproducible cheat-catch eval run in CI with no key. Feedback welcome — especially new cheats it should catch.

https://github.com/Saivineeth147/skeptic

1 reply

reacted to salma-remyx's post with 🔥 7 days ago

Post

3573

Nobody can be the originator of every great idea. And the rate of innovation in AI is difficult to track in detail.

But the next great idea for your codebase may have been described in open-science publications like arXiv.
Who among us can keep up with reading, smoke testing, adapting every viral method in the feed? (not me)

And what about the long-tail of results that you won't discover for another 6 months? That's what we built Outrider for!

Now I get alerts when candidate methods look like a good fit for the engineering challenges I'm addressing in my application.
I don't waste time running down dead end leads. Instead, I choose to invest my attention into ones which have earned it with evidence.

Try it yourself: https://studio.remyx.ai
Read the docs: https://docs.remyx.ai

reacted to ProCreations's post with 👀 7 days ago

Post

294

Grug 27b is being trained, it should hopefully be a lot better and more coherent. Follow me on X, https://x.com/sshthedev?s=11 When I reach 100 followers maybe Grug 100b 👀

1 reply

reacted to appvoid's post with 🔥 7 days ago

Post

3327

Two big projects are open sourced soon. Get ready...

11 replies

John Smith PRO

AI & ML interests

Recent Activity

Organizations

John6666's activity