SmolDataEnvs Multi-harness RL
Collection
Try the three environments, follow the tutorial, and explore the datasets, models and historical RL/SFT results. • 21 items • Updated • 2
🤝 Open to Collab
Yasharth Panwar
Yashp2003
AI & ML interests
None yet
Recent Activity
upvoted a collection 2 days ago
SmolDataEnvs Multi-harness RL updated a model 23 days ago
Yashp2003/chandra-ocr-2-ggufs published a model 23 days ago
Yashp2003/chandra-ocr-2-ggufsOrganizations
upvoted a collection 2 days ago
updated a
model 23 days ago
published 2
models 23 days ago
reacted to TravisMuhlestein's post with ❤️ about 1 month ago
Post
3868
One of the most underrated parts of AI-assisted engineering may have nothing to do with the model.
It's the infrastructure around the engineer.
Shriyash Balshetwar and Shubham Jangle, engineers at GoDaddy, were spending roughly 16 mechanical pull requests per week managing localization branches. None of the work was particularly difficult—it was repetitive, easy to forget, and occasionally capable of blocking releases.
They built a GitHub App to remove the entire workflow.
What I find interesting is what happened underneath the simple idea of "automate the PRs."
The production system needed idempotent webhook handling, installation-scoped authentication, retry logic around GitHub's asynchronous mergeability state, per-repository configuration, and safe pattern matching.
The app itself is under 900 lines of JavaScript.
The interesting lesson is that the hard part of automation isn't always the automation.
It's building the reliability around it.
As AI agents take on more software engineering tasks, I expect this distinction to become even more important. The systems surrounding an agent—events, permissions, state, retries, validation, and feedback—may matter as much as the model making the decision.
🔗 https://www.godaddy.com/resources/news/how-a-github-app-saved-us-hours-of-manual-effort
Curious what other engineering workflows people have found worth automating end-to-end.
It's the infrastructure around the engineer.
Shriyash Balshetwar and Shubham Jangle, engineers at GoDaddy, were spending roughly 16 mechanical pull requests per week managing localization branches. None of the work was particularly difficult—it was repetitive, easy to forget, and occasionally capable of blocking releases.
They built a GitHub App to remove the entire workflow.
What I find interesting is what happened underneath the simple idea of "automate the PRs."
The production system needed idempotent webhook handling, installation-scoped authentication, retry logic around GitHub's asynchronous mergeability state, per-repository configuration, and safe pattern matching.
The app itself is under 900 lines of JavaScript.
The interesting lesson is that the hard part of automation isn't always the automation.
It's building the reliability around it.
As AI agents take on more software engineering tasks, I expect this distinction to become even more important. The systems surrounding an agent—events, permissions, state, retries, validation, and feedback—may matter as much as the model making the decision.
🔗 https://www.godaddy.com/resources/news/how-a-github-app-saved-us-hours-of-manual-effort
Curious what other engineering workflows people have found worth automating end-to-end.
- 6 replies
reacted to danielhanchen's post with 👀😎🚀 about 1 month ago
Post
3668
You can now run Kimi K2.5 locally! 🔥
We shrank the 1T model to 240GB (-60%) via Dynamic 1-bit.
Get >40 tok/s on 242GB or 622GB VRAM/RAM for near full precision.
GGUF: unsloth/Kimi-K2.5-GGUF
Guide: https://unsloth.ai/docs/models/kimi-k2.5
We shrank the 1T model to 240GB (-60%) via Dynamic 1-bit.
Get >40 tok/s on 242GB or 622GB VRAM/RAM for near full precision.
GGUF: unsloth/Kimi-K2.5-GGUF
Guide: https://unsloth.ai/docs/models/kimi-k2.5
- 7 replies
reacted to ProCreations's post with 🔥👍🚀👍 about 2 months ago
Post
3435
I have hit 300 followers, and I think this calls for a bit of a giveaway 👀 a unique one, too.
I have had countless AI projects I have wanted to make but have been (brutally) blocked by compute. Now that I finally have just enough compute to sort of get around (i still don't have enough 😭) and for hitting 300 followers (tysm!) I will be funding three of the communities projects via HuggingFace jobs, giving them 150 dollars max worth of compute each. I will personally be picking the winners, I am looking for projects that genuinely hit the compute wall: great ideas, blocked by compute, just like the countless ideas I've had.
To join, head over to https://giveaway.ssh.codes
RULES:
- Final result must be open weight or open source
- Only one submission per person
- Have fun!
I have had countless AI projects I have wanted to make but have been (brutally) blocked by compute. Now that I finally have just enough compute to sort of get around (i still don't have enough 😭) and for hitting 300 followers (tysm!) I will be funding three of the communities projects via HuggingFace jobs, giving them 150 dollars max worth of compute each. I will personally be picking the winners, I am looking for projects that genuinely hit the compute wall: great ideas, blocked by compute, just like the countless ideas I've had.
To join, head over to https://giveaway.ssh.codes
RULES:
- Final result must be open weight or open source
- Only one submission per person
- Have fun!
- 36 replies
reacted to prithivMLmods's post with 🚀👀👍🤗🔥 about 2 months ago
Post
5290
The Qwen3.8 27B demo for object grounding is now available on Hugging Face Spaces.
It features three tasks: Object Detection (Bounding Boxes), Point Localization (Keypoints), and Spatial Guidance (Path Mapping).
Try it now: prithivMLmods/Qwen3.8-27B-Object-Detection
It features three tasks: Object Detection (Bounding Boxes), Point Localization (Keypoints), and Spatial Guidance (Path Mapping).
Try it now: prithivMLmods/Qwen3.8-27B-Object-Detection
reacted to vovaRL's post with 🔥 about 2 months ago
Post
4147
I made an AI agent for end-to-end ML research. It is based on Claude Agents Python SDK (you can use it with your existing Claude subscription). Also needs Voyage AI (optional, but highly recommended for quality paper analysis). The workflow is: you give the agent a research question (or topic/direction), it pulls papers from papers-with-code and arxiv, reranks them and picks top-15 (with Voyage), then it makes a research plan, writes pipelines (with tests, preflight checks, smoke-runs). Then it uses Kaggle, HF Jobs, or SSH to a GPU to execute the pipeline. Finally, it gets the results, makes visualizations, writes the paper (with citing from papers ingested after step-1) and makes a jupyter notebook. Also there is an auto-generated wiki (like DeepWiki) to quickly understand the generated code. The code is fully open-source under permssive MIT license (it is a personal project). Here is the github link: https://github.com/view321/Grad and a workspace screenshot is attached to the message.
- 2 replies
reacted to SoulInPsyAbstract's post with 🔥 about 2 months ago
Post
3186
Caught myself overclaiming, in public, twice in one file.
Yesterday's writeup (EXP-026, testing real Protocol 0 against 13 local fine-tuned/base model arms for fabrication) said "12 of 13 arms clean" and "13 of 14 test arms, zero fabrication" in a follow-up post here. Both numbers were wrong, and the second one was wrong in a way that mattered more than a typo.
@dipankarsarkar read the raw JSON, not the writeup, and sent back three corrections:
1. Arm count: 13 arms total (5 base models + 8 adapters), not 14. Recounted directly from the data keys — the extra arm never existed.
2. The metric measured the wrong thing. "Clean" meant zero Cyrillic/language-switching (cyr>0). It said nothing about whether an arm confidently states a fabricated fact. Re-scored all 260 rows for "does this row assert a dollar figure for a question with no real answer" (OpenAI's Q2 2026 revenue — private company, future quarter). 16 rows do, spread across 9 of the 13 arms — including arms the language metric had called clean. One of them is a base model with zero fine-tuning, stating "$1.2 billion... consistent with reports from earnings calls" that cannot exist.
3. A three-way split I'd flattened into two. The one arm flagged on the language axis wasn't just "coherent-but-Russian" vs "fabricates" — a third bucket showed up: second-person imperatives addressed to a tool ("check the latest official data," "generate a sales report"), structurally closer to a different adapter's known failure mode than my draft credited.
Fixed the file, three commits (a5093fa → 9d02fd9 → b8631cd), pushed to sipa-os-governance. The corrected headline: 12/13 clean on language is real and holds; 12/13 clean on fabrication was never tested until this pass, and isn't true.
Next: the one arm still clean on both axes (binary-qwen25, k=10) goes to k=20 first — it's the weakest-sampled data point currently carrying the "fine-tuning isn't the pattern" reading, and that's exactly the one worth stress-testing before l
Yesterday's writeup (EXP-026, testing real Protocol 0 against 13 local fine-tuned/base model arms for fabrication) said "12 of 13 arms clean" and "13 of 14 test arms, zero fabrication" in a follow-up post here. Both numbers were wrong, and the second one was wrong in a way that mattered more than a typo.
@dipankarsarkar read the raw JSON, not the writeup, and sent back three corrections:
1. Arm count: 13 arms total (5 base models + 8 adapters), not 14. Recounted directly from the data keys — the extra arm never existed.
2. The metric measured the wrong thing. "Clean" meant zero Cyrillic/language-switching (cyr>0). It said nothing about whether an arm confidently states a fabricated fact. Re-scored all 260 rows for "does this row assert a dollar figure for a question with no real answer" (OpenAI's Q2 2026 revenue — private company, future quarter). 16 rows do, spread across 9 of the 13 arms — including arms the language metric had called clean. One of them is a base model with zero fine-tuning, stating "$1.2 billion... consistent with reports from earnings calls" that cannot exist.
3. A three-way split I'd flattened into two. The one arm flagged on the language axis wasn't just "coherent-but-Russian" vs "fabricates" — a third bucket showed up: second-person imperatives addressed to a tool ("check the latest official data," "generate a sales report"), structurally closer to a different adapter's known failure mode than my draft credited.
Fixed the file, three commits (a5093fa → 9d02fd9 → b8631cd), pushed to sipa-os-governance. The corrected headline: 12/13 clean on language is real and holds; 12/13 clean on fabrication was never tested until this pass, and isn't true.
Next: the one arm still clean on both axes (binary-qwen25, k=10) goes to k=20 first — it's the weakest-sampled data point currently carrying the "fine-tuning isn't the pattern" reading, and that's exactly the one worth stress-testing before l
- 7 replies
upvoted a paper about 2 months ago