Rendered at 06:58:01 GMT+0000 (Coordinated Universal Time) with Cloudflare Workers.
coredog64 11 hours ago [-]
Good on these folks for doing something useful with Trainium. I'm always surprised that the Tranium ecosystem is as poor as it is. Amazon's market cap is nearly $3T and they can't throw a couple L6s and a modest token budget at this? Making Trainium work as seamlessly as possible as an Nvidia substitute is a clear flywheel play.
sanchitmonga22 5 hours ago [-]
facts! Trn2 and Trn3, is equivalent of Hopper and Blackwell, but the software support is poor!
We want to build the best engine in the open, welcoming any contributions!
bookernath 13 hours ago [-]
Tokens per minute is a strange metric, I usually see tok/s. A per-second figure would be more accessible in comparing this to alternatives.
11 hours ago [-]
kevo1ution 12 hours ago [-]
curious how the token/$ on trainuim compare to nvidia chips and amd. is the future "chipless" inference?
sanchitmonga22 5 hours ago [-]
We are able to prove out better token/$ for Trn2 vs H200!
There is a reason why Anthropic and OpenAI are choosing to run their inference on Trainium, it's highly gated right now, but we're trying to add support in the open and for open models!
badmonster0 13 hours ago [-]
how does this compare with sglang
shubham2802 11 hours ago [-]
They don't have support running trn2 afaik especially no glm 5.3 flash . they did release an update -5 days back but again its for trn3 nothing trn2
We want to build the best engine in the open, welcoming any contributions!
There is a reason why Anthropic and OpenAI are choosing to run their inference on Trainium, it's highly gated right now, but we're trying to add support in the open and for open models!