In a Training Loop 🔄

John Smith PRO

John6666

John6666cat

AI & ML interests

None yet

Recent Activity

liked a model about 13 hours ago

Sorihon/MN-Violet-Lotus-12B-Heretic

liked a model about 13 hours ago

mradermacher/MN-Violet-Lotus-12B-Heretic-i1-GGUF

liked a model about 13 hours ago

mradermacher/MN-Violet-Lotus-12B-Heretic-GGUF

View all activity

Organizations

reacted to PhysiQuanty's post with 🔥 about 17 hours ago

Post

1220

🧬 You can now find out whether your cognitive soulmate has already existed among 50k anonymized profiles ✨

SpiceeChat/Check-If-Your-Soulmate-Has-Already-Existed
SpiceeChat/OkCupid-59k-Anonymized-Profiles
https://dating-fatigue.com/

You seek them: 79.7% | They may seek you: 84.1% (coming soon)

🔥 Powered by open source and too much coffee 🔥

reacted to AxionLab-official's post with 👀 about 17 hours ago

Post

104

We RELEASED!

SupraLabs just released our 50M model!
Base, Instruct Weights are there, you can use!

You can check blog to more informations!(Writing blog yet!)

reacted to Shrijanagain's post with 🚀 about 17 hours ago

Post

1245

We are pleased to announce that the W-IMG Vision Dataset infrastructure is officially live.

The complete asset infrastructure is now accessible on Hugging Face for internal validation and architecture scaling targets.

Dataset Endpoint - sKT-Ai-Labs/W-IMG

#SovereignAI #ComputerVision #MachineLearning #OpenSource

reacted to Abdourakib's post with 👀 about 17 hours ago

Post

🧠 Just shipped ML Research Assistant on HuggingFace Spaces!

An AI research assistant powered by Qwen2.5-7B with 17 auto-routing tools. Ask in plain English — it figures out the right tool automatically.

🔧 What it can do:
• Compare LLMs on real benchmarks (MMLU, GSM8K, HumanEval, HellaSwag)
• Search ArXiv papers and find GitHub implementations
• Explore model benchmarks with real F1 and Accuracy scores
• Live AI news feed — click any headline to ask the assistant about it
• Web search, code generation, HuggingFace model and dataset search, and more

💡 Try these:
→ compare gpt-4o vs claude 3.5 sonnet on coding
→ show me the overall llm leaderboard
→ find papers with code for transformer attention
→ trending ml repos on github

⚙️ Built without LangChain — I wrote the router from scratch so the model only gets called once per query. That matters on ZeroGPU where GPU time is limited.

⚠️ Still a v1 — handles one query at a time, cold start delay on first load. Multi-query support and streaming are coming next.

🔗 Abdourakib/ml-research-assistant
📁 https://github.com/Abdrakib/ml-research-assistant

Would love feedback from the community — what would you add?

reacted to kanaria007's post with 👀 about 17 hours ago

Post

105

✅ Article highlight: *Receipted World Simulation Engine* (art-60-156, v0.1)

TL;DR:
This article treats WorldSim as a *governance sandbox*.

A game already has the right shape for SI: explicit world state, discrete actions, computed effects, verification, observability, and replay. So instead of asking “is this match fair?” by vibes, WorldSim makes fairness, anti-cheat, replay fidelity, patch legitimacy, and tournament claims depend on a *receipted closed loop*.

Read:
kanaria007/agi-structural-intelligence-protocols

Why it matters:
• makes governance feel concrete and intuitive instead of abstract
• shows that “a game is SI with better UX”
• turns match fairness, replay fidelity, and anti-cheat into artifact-backed claims
• connects gameplay operations to broader SI ideas: determinism, monitoring, patch governance, publication discipline, and interop

What’s inside:
• world state as *content-addressed state* with state_ref, ticks, shards, and canonicalization
• separate *action ledgers* and *effect ledgers* so “what happened” is reconstructible
• pinned determinism + *replay receipts* for faithful replay claims
• anti-cheat framed as *adversary monitoring* with monitoring receipts
• balance patches as governed change objects with shadow apply and verification
• tournament/public statements as bounded published claims, not vibes

Key idea:
Do not say:

*“this match was fair,”*
*“this replay is faithful,”*
or *“this tournament result is official.”*

Say:

*“this result is backed by a receipted closed loop: state, actions, effects, replay, verification, publication policy, and the exact pins needed to make the claim admissible.”*

reacted to barakor's post with 👀 about 17 hours ago

Post

I’ve just published a new article on State16 about a question I think will become increasingly important for robotics, embodied AI, VLAs, and world models.
Physical AI is already moving from research demos into real robots, vehicles, drones, and industrial systems. But as these models begin to produce actions in the physical world, we need to ask a harder question:

What is the runtime guardrail layer that makes these systems auditable, certifiable, and operationally safe?

In the post, I discuss the rise of World Foundation Models, Vision-Language-Action models, closed-loop autonomy stacks, and the missing layer between black-box AI policies and physical actuation.

Would be very interested to hear how this community thinks about runtime evaluation, action authorization, physical consistency checks, and safety layers for embodied AI systems.

Read here:
https://state16.ai/articles/physical-ai-is-already-here

Would love feedback, criticism, and pointers to relevant models, datasets, or evaluation frameworks.

reacted to johko's post with 👍 about 17 hours ago

Post

One prompt, three answers - which model is from where?

johko/llm-blind-date

I built a little demo where you give three models (Apertus, Llama, Qwen3) the same prompt and in the end you have to guess which is which just based on their answers.

GIve it a try! ;)

reacted to danielhanchen's post with 🤗👍 about 17 hours ago

Post

1229

Qwen3.6 MTP is here! Run locally on 20GB RAM. ⚡️

MTP enables Qwen3.6 to generate ~1.4–2.2× faster with no accuracy change.

Qwen3.6-27B: unsloth/Qwen3.6-27B-MTP-GGUF
Qwen3.6-35B-A3B: unsloth/Qwen3.6-35B-A3B-MTP-GGUF
Guide: https://unsloth.ai/docs/models/qwen3.6#mtp-guide

reacted to RiverRider's post with 🔥 about 17 hours ago

Post

🧠 New Space: MindReader-NLA — ask a frozen LM what it's thinking, in plain English.

A trained Activation Verbalizer (~5–13M params, frozen backbone) over Qwen-2.5-7B, Llama-3.2-3B, and Gemma-2-2B. Three demos in one Space:

Playground — sample K verbalizations of the layer-L hidden state and score how well each reproduces the original activation when fed back through the same frozen model (raw + anisotropy-centred cosine FVE).

Live Thought Trace — stream a verbalization per token as the model writes, side-by-side with the generation.

Steer-by-Editing — edit the verbalized thought, project it back into hidden-state space, and watch the continuation change.

Runs on ZeroGPU. Try it: RiverRider/srt-nla-demo

Paper + code: https://github.com/space-bacon/SRT

reacted to Crownelius's post with 🔥 about 17 hours ago

Post

4179

Howdy,
CompactAI-O is launching a tiny Model Golf, and the winner walks away with $50 in RunPod credits. Monthly. Every month. Show up, build, somebody wins.

What it is

Build the best language model you can under 100 million parameters, with at least a 1028-token context window. That's it. Any architecture, any tokenizer, any training scheme you can dream up at 3am. The only catch is it's gotta be open source (MIT, GPL, Apache, AGPL) take your pick.

It scratches the same itch as a Kaggle comp without the dataset\leaderboard nonsense. No fixed benchmark to game. No llama.cpp compatibility hoops. If you wanna train a 50M-param MoE with five experts and a tokenizer built on cookbooks, you can do that. Nothing stopping you.

The rules are listed in the discord and on the organization page if you're interested.

Why $50????

It's symbolic. It ain't gonna make anyone rich. But it's enough to cover a weekend of GPU time, enough to keep enthusiasts coming back, and not so much that it pulls in people who are just there for the money. Enthusiasts build interesting things. Interesting things move the field forward. A little incentive. I'd do it for $50 lol.

How to join

First round opens soon. Landing page is here:

→ CompactAI-O/Tiny-model-golf

For questions or to swap ideas, the Discord's open:

→ https://discord.gg/y2jTct6Cxv

Excited to see what yall come up with. ♥

— Shane

8 replies

reacted to etemiz's post with 👀 about 17 hours ago

Post

https://dystopiabench.com/ has a 0.53 correlation to my leaderboard https://aha-leaderboard.shakespeare.wtf/

it is cool to see supporting leaderboards. 0.53 is not too high but still validates both works. my idea is alignment through 'beneficial knowledge for humans', yet still there is correlation!

reacted to salma-remyx's post with 🚀 about 17 hours ago

Post

4969

Just trained a 2B coding model to rank candidate AI/ML research ideas against the implicit preferences in a code repository's merge history.

The training data comes from a Gaussian Process fit on the accumulated dispositions in VQASynth, where each PR against a deployed project yields a pairwise comparison between the feature branch preferred and the baseline at main.

The GP scores candidate papers to synthesize preference pairs, and DPO with LoRA bakes the ranking pipeline into the model's weights.

After 1 epoch the model reaches 87.4% reward accuracy on the held-out eval split against 92.3% on training, consistent with learning the task without overfitting.

Now, I'm scaling the pipeline to thousands of repos for a generalization test.

Dataset: remyxai/mhpd-dpo-v0
Model: remyxai/mhpd-dpo-qwen3.5-2b-vqasynth
Substack: https://remyxai.substack.com/p/the-ai-pm

reacted to danieldk's post with 🔥 about 17 hours ago

Post

110

Two large changes in kernel-builder this week:

kernel-builder now links libstdc++ dynamically. To support a wide range of systems, we build against libstdc++ from manylinux_2_28 (EL 8 and later).

Following our Torch support policy that the current and previous Torch versions are supported, Torch 2.10 support was removed. We will soon also support the Torch stable ABI, so that it is possible to write kernels that support a large number of Torch versions.

reacted to kavyamanohar's post with 🔥 about 17 hours ago

Post

4489

Releasing Vividh-ASR — an open benchmark and models for Hindi and Malayalam ASR.

Vividh-ASR is built from public data, stratified by complexity:
→ Clean recordings
→ Noisy and accented speech
→ Spontaneous, conversational audio

Alongside the benchmark, we release:
→ Open models for Hindi and Malayalam
→ A training recipe with two counterintuitive choices that moved the needle
→ What failed, not just what worked

The stratified evaluation methodology transfers directly to any low-resource language setup — beyond Hindi and Malayalam.

Built at @adalatai , where we build speech tech for Indian courts. This is our first open contribution back to the community. @janaab @Kush0610 @orgh0

Link: https://huggingface.co/blog/adalat-ai/vividh-benchmark

reacted to ManniX-ITA's post with 👀 3 days ago

Post

241

v1.1.0 was Claude + Ollama chat. Eight releases later the stack is a grounded research pipeline plus a local-first memory layer; the token crunch is operational now, not a quality wall.

🚀 claude-hooks v1.8.3 — highlights since v1.1.0.

🧠 /consultants v2 — agentic council, matured.
🛠 tool_executor — PLAN→REPORT lane runs read_file / grep / glob over the codebase before the researcher speaks; claims grounded in tool output, not vibes.
✍️─ coder — sandboxed write_file role with per-language model routing (50KB/file, 1MB/lane caps).
🛡️ CitationLinter — three-layer verifier at the researcher boundary; every path:line claim checked against an mtime-cached code_graph. Catches fabricated filenames before they launder through critics + synthesizer.

💾 M14 cross-session memory (default on).
LangGraph BaseStore wired across four namespaces: research / tool_results / project / user. Per-namespace TTL: research=30d, tool_results=24h, project+user=forever. Hourly Caliber-style distillation reaper summarizes
expiring research into the durable project namespace BEFORE deletion — episodic → semantic, like human consolidation. Originals only dropped after a successful summary write.

🔁 sqlite_vec — full pgvector parity (v1.7).
Hybrid recall via RRF over vector cosine + BM25 (FTS5). KG surface: kg_create_entities / kg_add_observations / kg_create_relations / kg_search_nodes. Bundled sqlite-vec-mcp launcher went 3→8 tools so Cursor / Codex /
OpenWebUI / Claude Desktop share the same .db. Lazy schema migration carries v1.6.x dbs in place, non-destructive.

🧩 llamafile chat + embed (v1.4 + v1.5).
HyDE / reflect / consolidate / get-advice / consultants route to a daemon-supervised local llamafile via the llamafile:// model prefix. Multi-instance LRU, per-label idle reap, sticky CPU fallback. Stack runs
offline now.

🐧 Linux / macOS / Windows. PostgreSQL OR SQLite. Local OR cloud LLMs.

🔗 github.com/mann1x/claude-hooks

1 reply

reacted to Doradus-AI's post with 👍 3 days ago

Post

184

Tonight we validated a small upstream vLLM fix that brings GLM-5.1-REAP-478B back into our consumer-Blackwell rotation pool.

Sleep/wake on 4× RTX PRO 6000 (SM_120) had a CuMemAllocator race that retired GLM in April: cuMemUnmap runs synchronously from the host the moment a pool-backed tensor's refcount hits zero, but kernels can still be in flight against that storage, accumulating CUDA_ERROR_ILLEGAL_ADDRESS,
engine eventually unrecoverable.

vllm-project/vllm#43020 is a one-line torch.cuda.synchronize() at the top of _python_free_callback. Steady-state inference unaffected (only cumem
frees pay the cost).

We caught the unpatched bug live during validation:

CUDA Error: invalid argument at /build/vllm/csrc/cumem_allocator.cpp:146

That's the exact failure class #43020 fixes. With it bind-mounted in: Q3.6-27B sleep/wake cycle clean (25.8 GiB VRAM released on /sleep level=1,
engine alive, post-wake chat coherent), GLM 30-request stress test 30/30 PASS, 0 CUDA errors. Back into rotation.

Side win: we're also submitting a generic Triton autotune shmem-budget helper upstream that replaces hand-rolled check_shared_mem() ? [64,128] :
[32,64] bucket switches with per-config precision via Triton's existing prune_configs_by={"early_config_prune": ...} hook. Zero change to the
H100/H200 fast path. Submitted: vllm-project/vllm#43047

Full writeup with byte math + stress-test logs + the bind-mount overlay pattern: https://doradusresearch.ai/blog/sleep-mode-on-blackwell-part-2/

Hardware: 4× NVIDIA RTX PRO 6000 Blackwell Workstation Edition (SM_120, 95 GiB per GPU, 101 KiB per-block opt-in shmem).

Image stack documented in the writeup!

reacted to kanaria007's post with 👀 3 days ago

Post

218

✅ Article highlight: Honest Benchmarking for Governed Intelligence Platforms (art-60-241, v0.1)

TL;DR:
This article argues that benchmark results should be published as bounded observations, not inflated into platform claims.

A governed benchmark should not quietly turn “we measured this result under these conditions” into “therefore this platform is more governed, safer, or more production-ready.” Honest benchmarking separates reproducibility, comparability, and disclosability—and keeps benchmark outcomes distinct from stronger governance or platform-readiness claims.

Read:
kanaria007/agi-structural-intelligence-protocols

Why it matters:
• prevents benchmark scores from being laundered into governance-readiness claims
• distinguishes reproducible results from truly comparable rankings
• makes public benchmark language respect disclosure floors and evidence class
• gives a clean way to publish strong numbers without overclaiming what they mean

What’s inside:
• the separation between reproducibility, comparability, and disclosability
• the rule that a benchmark result is not the same thing as a platform claim
• a benchmark disclosure profile that sets the publication floor
• a governed benchmark pack that binds runtime, toolchain, policy surface, evidence class, and results
• a comparability declaration and benchmark publication report that state what public reading is actually supportable

Key idea:
Do not say:

“we ranked higher, therefore we are better governed.”

Say:

“this governed benchmark pack produced these results under this disclosed runtime, toolchain, policy, and evidence surface; this comparability declaration defines what we are and are not fairly comparable to; and this publication report states exactly what public reading is supportable without inflating benchmark observations into stronger platform claims.”

reacted to tomaarsen's post with 🔥🤗 3 days ago

Post

337

🤗 Announcing the Ettin Reranker family: six new state-of-the-art CrossEncoder rerankers for search from 17M to 1B parameters, plus the full training data and the ~150-line recipe. Built on the Ettin ModernBERT encoders, Apache 2.0. Details:

All six were trained with the same single-stage pointwise MSE distillation recipe, with mixedbread-ai/mxbai-rerank-large-v2 (1.54B) as the teacher. Only the learning rate and per-device batch size change between sizes. The 1B student matches the teacher within 0.0001 NDCG@10 on MTEB(eng, v2) Retrieval, the 150M is the strongest reranker I tested in the under-600M range, and the 17M beats the 33M ms-marco-MiniLM-L12-v2 by +0.051 NDCG@10 at roughly half the parameter count.

Speed matters as much as quality for a reranker, since it determines whether the model fits the latency budget between retrieval and showing results. Our 17M is the fastest reranker in the whole comparison at 7517 pairs/sec on an H100. Our 150M runs 2.3x faster than the two other 150M ModernBERT-base rerankers (gte-reranker-modernbert-base and granite-embedding-reranker-english-r2) because the modular Transformer module propagates unpadded inputs through every layer rather than just the FA2 attention kernel. And our 1B is 2.4x faster than its 1.5B teacher while matching it on quality.

I bootstrapped the training recipe with the new train-sentence-transformers Agent Skill shipped in Sentence Transformers v5.5.0. Install it with hf skills add train-sentence-transformers --claude and ask Claude Code (or Codex / Cursor / Gemini CLI) to fine-tune a SentenceTransformer, CrossEncoder, or SparseEncoder model on your data.

I wrote a blog post walking through usage, results across six embedder pairings, the speed story, and the complete training script. Check it out, or just point your Agent to the URL:

https://huggingface.co/blog/ettin-reranker

Collection: https://huggingface.co/collections/cross-encoder/ettin-rerankers

John Smith PRO

AI & ML interests

Recent Activity

Organizations

John6666's activity