28 lines
734 B
Markdown
28 lines
734 B
Markdown
|
---
|
||
|
|
language: en
|
||
|
|
pipeline_tag: text-generation
|
||
|
|
tags:
|
||
|
|
- mlx
|
||
|
|
- meshllm
|
||
|
|
- parity
|
||
|
|
- same-origin
|
||
|
|
- olmo
|
||
|
|
library_name: mlx
|
||
|
|
license: apache-2.0
|
||
|
|
---
|
||
|
|
|
||
|
|
# OLMo-7B-Instruct-hf Parity BF16 MLX
|
||
|
|
|
||
|
|
Same-origin parity artifact derived from `allenai/OLMo-7B-Instruct-hf`.
|
||
|
|
|
||
|
|
This repo contains the high-fidelity `bf16` MLX artifact used for mesh-llm backend parity validation against the corresponding GGUF artifact.
|
||
|
|
|
||
|
|
Accepted local validation status:
|
||
|
|
|
||
|
|
- Exact prompts: MLX now tracks GGUF on the semantic probe set after prompt-render parity fixes, with shared family-level drift on strict one-word canaries
|
||
|
|
- Behavior smoke: 0 flagged prompts out of 80 on the MT-Bench-derived harness
|
||
|
|
|
||
|
|
Paired GGUF repo:
|
||
|
|
|
||
|
|
- `meshllm/olmo-7b-instruct-hf-parity-f16-gguf`
|