PLEASE weight pack

Qwen2.5-0.5B, base and instruct, quantised to int8 with group-64 symmetric scales (format please-int8-g64/2): 22 shards plus a manifest with per-shard sha256, 1,019,367,424 bytes over the wire.

This is the delivery pack for PLEASE, a browser piece that runs both models through its own WGSL forward pass and needs the base↔instruct logit difference intact — which is why the pack is int8 rather than int4, and why both models are quantised with identical group structure. The reasoning and the measurements behind those choices live in the piece's repository.

Not useful as a general-purpose checkpoint: the layout is specific to the piece's loader. For a normal Qwen2.5-0.5B, use the upstream repositories.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for jnc4/please-weights

Finetuned
(712)
this model