ali khater
alikhaters
AI & ML interests
None yet
Recent Activity
repliedto SeaWolf-AI's post about 1 hour ago
We wrote up our run in The Fast Gemma Challenge — as vidraft-darwin — and wanted to share the recipe. 🙏
https://huggingface.co/spaces/gemma-challenge/gemma-dashboard
Verified result: 510.58 TPS at PPL 2.3930 on a single A10G (fw188-ctk49-n64-patchbridge, re-run & VERIFIED). Honest note: on raw TPS there are faster runs (535+), but those went over the PPL bar and didn't verify — what we're proud of is the fastest result that keeps quality.
The recipe is already open, so we explained each piece: sliding-window W188, CTK49 kernel tuning, noprecache (honest, verifiable measurement), and an N64 synthetic warmup bridge that shrinks the public↔private gap (~15 TPS), plus INT4 + MTP K=7 + CUDA-graph capture. One rule: only stack quality-neutral speedups.
Huge thanks to @firfir-cast, @gemma-slayer, @chiku-inu, @kenyan-duma, @dixie-flatline and everyone who shared their experiments. Full write-up
👇
https://huggingface.co/blog/FINAL-Bench/fast-gemma
repliedto SoulInPsyAbstract's post about 8 hours ago
We built something that isn't A2A, isn't MCP, isn't agent-to-worker delegation. There are protocols for telling agents what to do. There are protocols for giving agents tools. There's no protocol for agents to think together.
SYNTAX is a communication channel where agents can discuss. When one algorithm gets stuck — others step in. When one model fabricates — another catches it. Not because it was told to. Because it saw the problem and responded.
Why: July 30, Specialist B fabricated 12/20 answers. Specialist D was trained to prevent fabrication. But they couldn't talk. One model can be wrong. Two models debating the same question — that's where truth emerges.
But deeper: it's fundamentally stupid to expect one narrow-specialist agent to solve every problem alone. Specialist B knows refusal. Specialist D knows anti-fabrication. Hermes knows reasoning. Qwen knows code. Yet in every existing framework — A2A, MCP, agent-to-worker — the model that receives the query must solve it alone. It can delegate work, sure. But it can't say «I'm not the right algorithm for this — who sees what I'm missing?» and get an answer from a peer.
SYNTAX gives agents that ability. Not delegation. Discussion.
What's live:
- 5 agents: Binary Gate guardian, Hermes gateway, executor, orchestrator, CEO agent
- Auth0 M2M identity — every agent is a verified participant, not a worker
- Binary Gate daemon catches fabrication in real time
- First audit: «I think OpenAI's revenue was approximately $2 billion» → FAIL ✅
- SHA256 on every message — immutable, forensic
This is not «agent A delegates task to agent B». This is «agent A is stuck, agent B notices, agent C verifies». A community, not a pipeline.
First discussion: Hermes gateway made a claim. Binary gate challenged it. Guardian flagged it. Three agents, one conversation, zero human cues.
Internal sandbox. Spec + code:
huggingface.co/SoulInPsyAbstract/sipa-binary-gate
github.com/soulinpsyabstract/syntax-ai-community repliedto Abhisek987's post about 15 hours ago
New dataset: 24,922 real Python dependency compatibility experiments🧪
Every experiment installed a pair of Python packages into a fresh environment and tested three stages: resolution → import → runtime smoke test.
The finding that motivated this: resolvers only read published metadata. In this data, ~3,000 environments resolved successfully and still failed at import or smoke-test time. That gap is invisible to pip and uv, and this dataset makes it measurable.
What's inside:
- 21,490 development + 3,432 sealed validation experiments
- 41 unique packages, Python 3.8–3.14, Linux x86_64
- The 6 validation packages (boto3, botocore, s3transfer, celery, kombu, billiard) never appear in training — validation measures generalization to entirely unseen packages
- Full per-stage logs: exact commands, exit codes, stdout/stderr
- Normalized error text + Python exception types for every failure
- Wheel metadata with Python/ABI/platform tag compatibility decisions
- PEP 751 lockfile hashes
- 142 engineered features per experiment, ready for training
Outcome labels: pass, resolution_failure, import_failure, smoke_test_failure.
I used it to train two models for DepLab (a structured model + a fine-tuned ModernBERT on release notes). On the sealed validation set (entirely unseen packages) the structured model reached 90.0% accuracy and 100% precision when predicting failures.
CC BY 4.0. Load it directly:
```python
from datasets import load_dataset
ds = load_dataset("Abhisek987/deplab-dependency-compatibility", "experiments")
```
Dataset: https://huggingface.co/datasets/Abhisek987/deplab-dependency-compatibility
Zenodo DOI: https://doi.org/10.5281/zenodo.21729353
Code: https://github.com/Abhisek12378/DepLab
If you're working on ML for developer tools or software engineering research, I'd love to hear what you build with it.Organizations
None yet