compute: Current release card for a model-directory release (Qwen3.8-27B) - #1377
Merged
Conversation
The serving release moved from Qwen3.6-35B-A3B (one GGUF file pinned by
MODEL_SHA256) to Qwen3.8-27B NVFP4, a Hugging Face model directory pinned
by commit plus a sha256 per weight shard. das now exposes two more fields
on the release object:
modelDirSha256 "file=sha256,file=sha256" for the shards; null for a GGUF
runtimeEnv extra env the runtime container needs (CTX, MODEL_NAME,
TOK_REPO); null when the image defaults suffice
For a directory release modelSha256 is null and modelFile is
"<hf repo>@<hf commit sha>".
ServingRelease grows both fields and modelSha256 becomes nullable. The
card's `.env` and `docker run` builders share one pair list: a GGUF
release still emits MODEL_SHA256; a directory release emits
MODEL_DIR_REPO / MODEL_DIR_REVISION (split from modelFile at the last '@')
/ MODEL_DIR_SHA256, then every runtimeEnv key — exactly what
docker-compose.miner.yml passes to docker/sparkinfer-entrypoint.sh, which
refuses to start on a digest mismatch. The card shows "Model directory"
(copyable) and a "Shard SHA-256" block (one line per shard, copies the
joined string) for a directory release, "Model file" + "Model SHA-256" for
a GGUF, and a "Runtime env" field whenever the release carries one.
The dev mock becomes the 27B directory release with its real image digest,
commit and shard digests, and the fleet figures follow the 27B (~99 tok/s
single-stream, ~338 aggregate, ≈$0.58 per million output tokens).
Claude-Session: https://claude.ai/code/session_01VjhStJbNW6WzxucTcwuw4y
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
The serving release moves to Qwen3.8-27B NVFP4 (entrius/gittensor#1758), a Hugging Face model directory pinned by commit + per-shard sha256 rather than one GGUF file. The Current release card now renders that shape and emits the matching miner
.env/docker runlines.ServingReleasegainsmodelDirSha256(file=sha256,…) andruntimeEnv({CTX, MODEL_NAME, TOK_REPO}), both nullable;modelSha256is nullable. Both come from two newserving_roundscolumns (entrius/gittensor-db PR) that das passes through (entrius/das-gittensor PR).repo@commit, copyable), Shard SHA-256 block (one per line, copies as the comma-joined string), Runtime env field;.envblock =RUNTIME_IMAGE,ATTEST_IMAGE,MODEL_DIR_REPO,MODEL_DIR_REVISION,MODEL_DIR_SHA256,CTX,MODEL_NAME,TOK_REPO— exactly whatdocker-compose.miner.yml/miner.env.exampleread. The docker command passes the same as-e.MODEL_SHA256=).buildRuntimeEnvPairs) feeds both blocks so they cannot drift.Order
Merge after gittensor-db (columns) and das-gittensor (fields); the card degrades to the GGUF look while
modelDirSha256is null, so merging early is harmless.Checks
tsc -b,eslint --max-warnings 0, prettier: clean.https://claude.ai/code/session_01VjhStJbNW6WzxucTcwuw4y