docs: add on-disk sizes for GLM 5.2 model variants - #728
Open
vincenzopalazzo wants to merge 1 commit into
Open
Conversation
The DeepSeek Flash variants already document their sizes and RAM class
in download_model.sh, but the GLM 5.2 entries only described the quant
layout. Measured on disk after download:
glm-antirez-iq2xxs ~211 GB (does not fully fit 128 GB RAM;
use --ssd-streaming on single-node)
glm-antirez-q2 ~262 GB (already documented)
glm-antirez-q4 ~434 GB (already documented)
glm-unsloth-q4 ~467 GB (11 shards)
Also annotate the README download list with the same sizes so a
quick glance is enough to pick a machine class.
vincenzopalazzo
added a commit
to vincenzopalazzo/local-ai
that referenced
this pull request
Aug 6, 2026
For science: iq2xxs is 211 GB (not 81 — that was ds4f-q2). SSD streaming loads an ~800B MoE in ~47 GiB resident, but CUDA GLM prefill fails silently on GB10 (host registration unsupported; Metal/ROCm-first streaming paths). PR antirez/ds4#728 adds the missing GLM sizes upstream.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
download_model.sh(iq2xxs was the only one missing a size; q2/q4 already had them)--ssd-streamingContext
Verified by downloading both
glm-antirez-iq2xxsandglm-antirez-q2onto a DGX Spark (GB10, 128 GB unified memory) while setting up a local bake-off. The missing size on iq2xxs made it easy to confuse with the DeepSeek Flashds4f-q2entry ("about 81 GB… recommended for 96 and 128 GB RAM machines"), which sits right above the GLM block in the help text.Also confirmed
--ssd-streaming --ssd-streaming-cache-experts 40GBloads iq2xxs successfully on a single Spark (~46 GiB resident: 0.94 model + 35 expert cache + 5.8 compressed KV + buffers).Test plan
bash -n download_model.sh(syntax)ls -laafter full HF download of both files