Baseten Inference Stack
The encoder, made fast
Six changes to the multimodal encoder and its request path. Hit ▶ Play on each one to watch the before, the after, and what it bought.
2.0×encoder end-to-end
1040 → 510 ms / 4K image
−41%payload per response
base64 gone from every hop
+10–20%throughput
pack overlaps the next forward
566 req/son cache hits
at 1.8 ms each
0.9994new-model parity
GLM-5.2-Vision, live in prod
First — where the time actually goes
282 ms per image, profiled honestly
Change 1 · #23973 · merged
The encoder becomes its own package
Change 2 · #21313 · merged
Double the encoder itself
Change 3 · #23921 · #23964 · #24400 · merged
Stop paying the base64 tax
Change 4 · #24032 · merged
Pack while the next image runs
Change 5 · #24403 · open — needs a merge
Free the blob the moment it's dead
Change 6 · #24454 · merged
Recount the tokens after the splice
The payoff · #24242 · merged
New model in 34 lines: GLM-5.2-Vision
Still open: #24403 (free blobs at dispatch) and #24540 (GLM encoder packaging). Everything else above is merged — states verified Jul 29.