Prabakaran Chandran PRO

prabakaranchandran

pracha98
prabakaranc98
prabakaranchandrantheds

AI & ML interests

Intelligence : Post Training, Reasoning, Agentic Systems, AI-Human-Alignment and Safety, Causal Inference, World Models, Control Systems, Reinforcement Learning, Vertical AI

Recent Activity

updated a collection 2 days ago

Baby Pooh: growing a Pooh character model

updated a model 2 days ago

prabakaranchandran/baby-pooh-46m-midtrain-r4b

published a model 2 days ago

prabakaranchandran/baby-pooh-46m-midtrain-r4b

View all activity

Organizations

prabakaranchandran 's collections 4

Baby Pooh: growing a Pooh character model

Experimental: a 46M LM grown from scratch on child-scale data, then mid-trained on A. A. Milne's Pooh world. Models and data.

prabakaranchandran/baby-pooh-46m

Text Generation • 46.2M • Updated 3 days ago • 502
prabakaranchandran/baby-pooh-46m-midtrain-r2

Text Generation • 46.2M • Updated 2 days ago • 403
prabakaranchandran/baby-pooh-46m-midtrain-r3

Text Generation • 46.2M • Updated 2 days ago • 405
prabakaranchandran/pooh-babylm-mix-220m

Viewer • Updated 3 days ago • 308k • 191

ex1-medbench-rl · MedCalc-Bench (experimental, WIP)

Experimental (WIP): Qwen3-0.6B on MedCalc-Bench Verified with GRPO and on-policy distillation (JAX). Not for clinical use.

prabakaranchandran/ex1-medbench-rl-medcalc-opd-lora-qwen3-0.6b

Text Generation • 0.6B • Updated 8 days ago • 42
prabakaranchandran/ex1-medbench-rl-medcalc-grpo-lora-qwen3-0.6b

Text Generation • 0.6B • Updated 8 days ago • 27

ex1-medbench-rl · MedAgentBench (experimental, WIP)

Experimental (WIP): Qwen3-0.6B as a FHIR agent on MedAgentBench; SFT warm-up then multi-turn GRPO. Not for clinical use.

prabakaranchandran/ex1-medbench-rl-medagent-sft-grpo-qwen3-0.6b

Text Generation • 0.6B • Updated 8 days ago • 32

ex1-medbench-rl (experimental, WIP)

Experimental (WIP) Qwen3-0.6B checkpoints: GRPO, on-policy distillation and agentic RL on MedCalc-Bench and MedAgentBench. Not for clinical use.

prabakaranchandran/ex1-medbench-rl-medcalc-opd-lora-qwen3-0.6b

Text Generation • 0.6B • Updated 8 days ago • 42
prabakaranchandran/ex1-medbench-rl-medagent-sft-grpo-qwen3-0.6b

Text Generation • 0.6B • Updated 8 days ago • 32
prabakaranchandran/ex1-medbench-rl-medcalc-grpo-lora-qwen3-0.6b

Text Generation • 0.6B • Updated 8 days ago • 27

Baby Pooh: growing a Pooh character model

Experimental: a 46M LM grown from scratch on child-scale data, then mid-trained on A. A. Milne's Pooh world. Models and data.

prabakaranchandran/baby-pooh-46m

Text Generation • 46.2M • Updated 3 days ago • 502
prabakaranchandran/baby-pooh-46m-midtrain-r2

Text Generation • 46.2M • Updated 2 days ago • 403
prabakaranchandran/baby-pooh-46m-midtrain-r3

Text Generation • 46.2M • Updated 2 days ago • 405
prabakaranchandran/pooh-babylm-mix-220m

Viewer • Updated 3 days ago • 308k • 191

ex1-medbench-rl · MedAgentBench (experimental, WIP)

Experimental (WIP): Qwen3-0.6B as a FHIR agent on MedAgentBench; SFT warm-up then multi-turn GRPO. Not for clinical use.

prabakaranchandran/ex1-medbench-rl-medagent-sft-grpo-qwen3-0.6b

Text Generation • 0.6B • Updated 8 days ago • 32

ex1-medbench-rl · MedCalc-Bench (experimental, WIP)

Experimental (WIP): Qwen3-0.6B on MedCalc-Bench Verified with GRPO and on-policy distillation (JAX). Not for clinical use.

prabakaranchandran/ex1-medbench-rl-medcalc-opd-lora-qwen3-0.6b

Text Generation • 0.6B • Updated 8 days ago • 42
prabakaranchandran/ex1-medbench-rl-medcalc-grpo-lora-qwen3-0.6b

Text Generation • 0.6B • Updated 8 days ago • 27

ex1-medbench-rl (experimental, WIP)

Experimental (WIP) Qwen3-0.6B checkpoints: GRPO, on-policy distillation and agentic RL on MedCalc-Bench and MedAgentBench. Not for clinical use.

prabakaranchandran/ex1-medbench-rl-medcalc-opd-lora-qwen3-0.6b

Text Generation • 0.6B • Updated 8 days ago • 42
prabakaranchandran/ex1-medbench-rl-medagent-sft-grpo-qwen3-0.6b

Text Generation • 0.6B • Updated 8 days ago • 32
prabakaranchandran/ex1-medbench-rl-medcalc-grpo-lora-qwen3-0.6b

Text Generation • 0.6B • Updated 8 days ago • 27