Sahil Chaudhary
@csahil28
Followers
5K
Following
1K
Statuses
442
RT @NousResearch: Introducing DeepHermes-3 Preview, a new LLM that unifies reasoning and intuitive language model capabilities. https://t.…
0
63
0
I understand the negative impact this has had on the open-source AI community. I’m committed to learning from these mistakes and hope this post-mortem adds clarity to what happened. Moving forward, I will be more careful and thorough in any of our releases and communications. I’m dedicated to rebuilding trust and contributing positively to the open source AI ecosystem.
6
3
129
I want to address the confusion and valid criticisms that this has caused in the community. I am currently investigating what happened that led to this and will share a transparent summary as soon as possible. There are two areas I’d like to address, which I am investigating: - First, I want to be clear that at no point was I running any models from other providers as the API that was being served on my compute — I’m working on providing evidence of this and understanding why people saw model behaviour such as using a different tokenizer, or completely skipping words like “Claude”. - Second, the benchmark scores I shared with Matt haven’t been reproducible so far. I am working to understand why this is and if the original scores I reported were accurate or a result of contamination / misconfiguration. I have a lot of work to do on both of these and am working on a full postmortem that I will share with the community. I’m sorry for the confusion this has caused and know that I’ve let the community down and lost trust. I still believe in the potential of the approach. My focus is on rebuilding trust through increased transparency. I’ll have more to share soon.
I got ahead of myself when I announced this project, and I am sorry. That was not my intention. I made a decision to ship this new approach based on the information that we had at the moment. I know that many of you are excited about the potential for this and are now skeptical. Nobody is more excited about the potential for this approach than I am. For the moment, we have a team working tirelessly to understand what happened and will determine how to proceed once we get to the bottom of it. Once we have all of the facts, we will continue to be transparent with the community about what happened and next steps.
82
20
320
RT @Yuchenj_UW: We @hyperbolic_labs now serve Reflection 70B by @mattshumer_ in FP16! 🤖🔥 > Use our API or playground to play w/ it > It’s…
0
30
0
@abacaj @ImSh4yy @mattshumer_ @GlaiveAI had some trouble with uploads so it's sharded into 2gb files, will upload fp16 with bigger shard size as well, for easier downloads
1
0
2
@hu_yifei @mattshumer_ @GlaiveAI We had a few issues when uploading and had to reduce shard size, will be uploading fp16 with fewer files soon
0
0
6
RT @altryne: This from @mattshumer_ and @csahil28 (@GlaiveAI ) is insane! A LLama 70B finetune that has reflection baked into it's weight…
0
2
0
RT @mattshumer_: I'm excited to announce Reflection 70B, the world’s top open-source model. Trained using Reflection-Tuning, a technique d…
0
1K
0
RT @NousResearch: Introducing 𝐇𝐞𝐫𝐦𝐞𝐬 𝟑: The latest version in our Hermes series, a generalist language model 𝐚𝐥𝐢𝐠𝐧𝐞𝐝 𝐭𝐨 𝐲𝐨𝐮. https://t.co/…
0
313
0