Hanze Dong
@hendrydong
Followers
329
Following
291
Statuses
184
Research Scientist @SFResearch | Reproducibility & interpretability of LLMs | Sampling Algorithm | Core author of LMFlow, RLHFlow, Iterative SFT/DPO (RAFT/GSHF)
Joined June 2011
RT @SFResearch: ⚡ Meet BOLT: A novel approach to develop long chain-of-thought reasoning in LLMs without relying on knowledge distillation…
0
30
0
😂😂😂
NVIDIA and CMU presents ASAP, which enables highly agile motions that were previously difficult to achieve! @Cristiano Siuuuuuuu!
0
0
1
RT @baohao_liao: Impressed by DeepSeek-R1 and o3? However, they are long-reasoning models, and generate >4k tokens quite often for hard que…
0
5
0
Many thanks to all the collaborators @baohao_liao @xyh6666 @LiJunnan0409 @c_monz @silviocinguetta @doyensahoo @CaimingXiong
1
0
4
Very interesting. Btw, as the population declines, industries focused on resource extraction will face pressure to shift towards innovations, crucial for improving labor efficiency and sustaining growth with less workforce. The system itself might have to break the equilibrium.
0
0
4
RT @JacobSteinhardt: In 2021, our research group released the MATH dataset. In the paper, we attribute the data to math contests released b…
0
13
0
RT @rosstaylor90: “Wait that can’t be right” in the wild Thank you internet anons for your service to LLM reasoning. We found you through…
0
19
0
RT @Ber18791531: It's exciting to see Kimi-k1.5 uses a similar RL training objective to our (response-level) OREO! The difference is they…
0
26
0
Not human users but other LLMs might subscribe to the pro version 🤣
insane thing: we are currently losing money on openai pro subscriptions! people use it much more than we expected.
0
0
1
Interesting
Tier 1: - Sequoia - Founders Fund - A16Z - YC Tier 2: - Benchmark - General Catalyst - Khosla - Lightspeed VP - Index - Kleiner Perkins - Caffeinated Capital - SV Angel - Tiger Global - First Round - Greenoaks - Accel - Bessemer - Greylock - USV - Paradigm - Homebrew - Form Cap - Menlo - Craft Tier 3 & beyond: - Everyone else Unranked: - There are some newer unranked funds that have a lot to prove that I wouldn’t include in Tier 3.
0
0
1
RT @danielhanchen: Cool things from DeepSeek v3's paper: 1. Float8 uses E4M3 for forward & backward - no E5M2 2. Every 4th FP8 accumulate…
0
256
0
RT @jiayq: In 2019 I had a chat with the DeepSeek team, in the hope of selling them an AI cloud solution. I was trying to convince them a f…
0
123
0
RT @deepseek_ai: 🚀 Introducing DeepSeek-V3! Biggest leap forward yet: ⚡ 60 tokens/second (3x faster than V2!) 💪 Enhanced capabilities 🛠 AP…
0
2K
0
RT @gm8xx8: Offline Reinforcement Learning for LLM Multi-Step Reasoning OREO (Offline Reasoning Optimization) is introduced to enhance the…
0
6
0