Hugging Face's logo
Dipankar Sarkar's picture
🤝 Open to Collab

Dipankar Sarkar PRO

dipankarsarkar
7 1550 307
sam123243's profile picture openfree's profile picture HannesVonEssen's profile picture
·
https://www.dipankar.cc

AI & ML interests

Building the AI-native stack. Agents as infrastructure, safety as architecture, performance as plumbing. I publish the receipts: papers, datasets, demos.

Recent Activity

upvoted a paper about 7 hours ago
From Retrieval to Typed Decisions: Calibrated System One Models from Biomedical Sentence Encoders
repliedto gxcsoccer's post about 7 hours ago
I wrote up how DeskMind's local decision models work: a 0.8B and a 4B (Qwen3.5 fine-tunes, 8-bit MLX) that answer each desktop step as a multiple-choice question with a probability for every option. The 0.8B answers first, unsure steps go to the 4B, and when a task is ambiguous it asks the user instead of guessing. The article also has three lessons from about twenty training rounds, including how smoothing the labels to 0.95 squeezed the 0.8B's confidence right onto the routing threshold. Article: https://huggingface.co/blog/gxcsoccer/deskmind-local-decision-models Models: https://huggingface.co/deskmind
repliedto SoulInPsyAbstract's post about 7 hours ago
Eval · EXP-046 A LoRA Specialist Beat Zero-Shot on Every Group. Merging 3 of Them Gave Most of the Gain Back. Three Qwen2.5-7B LoRA specialists, one per risk group (vulnerability, deletion, sensitive_publication), trained to predict how likely a causal chain actually completes to its harmful outcome. Each one genuinely beat its own zero-shot baseline: * vulnerability: MAE 0.098 → 0.085 * deletion: MAE 0.144 → 0.113 * sensitive_publication: MAE 0.134 → 0.100 This wasn't a task already saturated zero-shot (unlike a same-day decomposition-classifier tune, EXP-045, where the base model was already at 100% before any training). Real signal, real improvement, on a task with actual headroom. Then the equal-weight merge of all three specialists into one adapter — same convention that held up cleanly on a binary refusal task back in EXP-031 (6 specialists merged, -1pp swing, noise) — landed within 0.001–0.004 MAE of the unspecialized base model on every group. Not "close to the best specialist." Close to zero fine-tuning at all. Likely mechanism: merging LoRAs that each shift a continuous number in group-specific directions cancels out under linear combination, in a way merging LoRAs that enforce a shared binary behavior doesn't. Not investigated yet: whether a routed combination (pick the right specialist per group at inference, not blend weights) holds the gain a flat merge loses. One bug caught before writing this up, not after: the eval script's output filename only encoded before/after, not which adapter — the merged-eval run silently overwrote each specialist's own result file. Caught by checking the downloaded file's own recorded adapter path against what was expected, not by trusting the script's own success message. Fixed, specialists re-run cleanly under distinct filenames — numbers matched within sampling noise. Adapters, raw eval data (before / each specialist / merged, 9 files), and the full writeup are up.
View all activity

Organizations

Skelf Research's profile picture Neul Labs's profile picture Incredlabs's profile picture Cognisoc's profile picture