swarnaNLP Profile Banner
Swarnadeep Saha Profile
Swarnadeep Saha

@swarnaNLP

Followers
1K
Following
1K
Media
57
Statuses
622

Research Scientist @AIatMeta (FAIR) working on Reasoning. Past: @Google PhD fellow @uncnlp. Gooner.

Seattle, Washington
Joined May 2014
Don't wanna be here? Send us removal request.
@swarnaNLP
Swarnadeep Saha
2 months
Progress of AI is bottlenecked by the quality of evaluation, motivating the need for powerful and generalist LLM judges that can think and reason. Here's our latest paper, J1, on how to train such Thinking-LLM-Judges with RL. 🧵👇.
@jaseweston
Jason Weston
2 months
🚨 New paper 🚨.J1: Incentivizing Thinking in LLM-as-a-Judge via RL. - Converts judgement task into a verifiable one for both verifiable and non-verifiable prompts. Uses only synthetic pairwise data. - Optimizes thoughts, scores, and judgments using GRPO. - Outperforms all
Tweet media one
2
3
58
@swarnaNLP
Swarnadeep Saha
10 days
Check out our new paper where we compared offline and (Semi-)Online DPO with GRPO for post-training LLMs. This led to some interesting findings! 👇.
@jaseweston
Jason Weston
11 days
🌉 Bridging Offline & Online RL for LLMs 🌉.📝: New paper shows on verifiable & non-verifiable tasks:.- Online DPO & GRPO give similar performance. - Semi-online (iterative) DPO with sync every s steps (more efficient!) works very well also. - Offline DPO
Tweet media one
1
1
8
@swarnaNLP
Swarnadeep Saha
1 month
RT @dair_ai: 3. J1. Introduces a novel training approach for LLMs to act as evaluators (LLM-as-a-Judge) by explicitly incentivizing thought….
0
1
0
@swarnaNLP
Swarnadeep Saha
2 months
RT @johnschulman2: For people who don't like Claude's behavior here (and I think it's totally valid to disagree with it), I encourage you t….
0
40
0
@swarnaNLP
Swarnadeep Saha
2 months
RT @rohanpaul_ai: Evaluation of LLMs is difficult due to judge models using limited reasoning and suffering from biases. This paper propos….
0
1
0
@swarnaNLP
Swarnadeep Saha
2 months
RT @TheTuringPost: The freshest research of the week:. Our top 9:.▪️ Beyond 'Aha!'.▪️ J1: Incentivizing Thinking in LLM-as-a-Judge via Rein….
0
11
0
@swarnaNLP
Swarnadeep Saha
2 months
We're organizing the RAM 2 workshop at COLM 2025 (10 years after the first edition in NeurIPS 2015). Check out our Call of Papers on topics in Reasoning, Attention, and Memory.
@jaseweston
Jason Weston
2 months
🚨Announcing RAM 2 workshop @ COLM25 - call for papers🚨 .- 10 years on, we present the sequel to the classic RAM🐏 (Reasoning, Attention, Memory) workshop that took place in 2015 at the cusp of major change in the area. Now in 2025 we reflect on what's happened and discuss the
Tweet media one
Tweet media two
0
0
5
@swarnaNLP
Swarnadeep Saha
2 months
RT @chenxi_jw: Presenting new work: Thinking LLM-as-a-Judge via RL!. It’s been great fun working with @swarnaNLP, @jaseweston, @uralik1 and….
0
1
0
@swarnaNLP
Swarnadeep Saha
2 months
RT @NathanThinks: excellent work by @jaseweston & team—extending our "Generative Reward Models" work with RL (GRPO) to optimize LLM reasoni….
0
11
0
@swarnaNLP
Swarnadeep Saha
2 months
Check out our paper for more analysis and ablations, including:. - score distribution of Pointwise-J1 models. - different reward schemes. - different seed thinking prompts. - reward+thought lengths. Fun Fact: Even before the project started, we knew what we wanted to call it 😀
Tweet media one
Tweet media two
0
0
1
@swarnaNLP
Swarnadeep Saha
2 months
Next, Pointwise-J1 at 8B + 70B scales:. 1⃣ Mitigates position bias. 2⃣ Improves position-consistent accuracy. 3⃣ Reduces ties in pairwise judgments. Finally, test-time scaling of J1 leads to further improvements, for both Pairwise and Pointwise models, at both scales
Tweet media one
1
0
0
@swarnaNLP
Swarnadeep Saha
2 months
We tested J1 on 5 benchmarks w/ verifiable + non-verifiable + multilingual instructions at 8B+70B scales. First, Pairwise-J1 outperforms:. 1⃣ open + closed LLM judges . 2⃣ SOTA scalar + generative RMs. 3⃣ R1-distilled-Llama + o1-mini. 4⃣ a much larger R1 on non-verifiable tasks
Tweet media one
1
0
1
@swarnaNLP
Swarnadeep Saha
2 months
🧑‍🍳 J1 Recipe:. -- Generate synthetic preference pairs as training data for both verifiable+non-verifiable tasks. -- Train Pairwise-J1 using GRPO with verdict correctness+consistency rewards. -- Train Pointwise-J1 using GRPO with distant pairwise supervision + score-based rewards
Tweet media one
1
0
2
@swarnaNLP
Swarnadeep Saha
2 months
Excited to share that EvalPlanner is accepted to #ICML2025! . To make meaningful progress in AI, we need strong evaluators, and specifically those that can reason. Stay tuned for more updates, as we continue to make progress in this space! 😀.
@jaseweston
Jason Weston
5 months
💭🔎 Introducing EvalPlanner – a method to train a Thinking-LLM-as-a-Judge that learns to generate planning & reasoning CoTs for evaluation. Strong performance on RewardBench, RM-Bench, JudgeBench & FollowBenchEval. Paper 📄:
Tweet media one
5
11
82
@swarnaNLP
Swarnadeep Saha
2 months
RT @SomnathBrc: 𝐇𝐨𝐰 𝐜𝐚𝐧 𝐰𝐞 𝐩𝐞𝐫𝐟𝐞𝐜𝐭𝐥𝐲 𝐞𝐫𝐚𝐬𝐞 𝐜𝐨𝐧𝐜𝐞𝐩𝐭𝐬 𝐟𝐫𝐨𝐦 𝐋𝐋𝐌𝐬?. Our method, Perfect Erasure Functions (PEF), erases concepts from LLM repre….
0
35
0
@swarnaNLP
Swarnadeep Saha
3 months
RT @tesatory: Ten years ago in 2015 we published a paper called End-to-End Memory Networks (. Looking back, this pa….
0
117
0
@swarnaNLP
Swarnadeep Saha
3 months
RT @jaseweston: 🚨Multi-Token Attention🚨.📝: Attention is critical for LLMs, but its weights are computed by single….
0
148
0
@swarnaNLP
Swarnadeep Saha
5 months
RT @ArchikiPrasad: 🚨 Excited to share: "Learning to Generate Unit Tests for Automated Debugging" 🚨.which introduces ✨UTGen and UTDebug✨ for….
0
61
0
@swarnaNLP
Swarnadeep Saha
5 months
Complex evaluation is in parts planning and in parts reasoning. Hence, we trained an LLM to think before producing a judgment. My first work since joining this awesome team 😄.
@jaseweston
Jason Weston
5 months
💭🔎 Introducing EvalPlanner – a method to train a Thinking-LLM-as-a-Judge that learns to generate planning & reasoning CoTs for evaluation. Strong performance on RewardBench, RM-Bench, JudgeBench & FollowBenchEval. Paper 📄:
Tweet media one
4
15
75
@swarnaNLP
Swarnadeep Saha
5 months
RT @mohitban47: 🎉 Congrats to the awesome students, postdocs, & collaborators for this exciting batch of #ICLR2025 and #NAACL2025 accepted….
0
31
0