mirror of
https://github.com/Dynamis-Labs/spectralquant.git
synced 2026-10-19 00:39:46 -05:00
Ashwin GopinathandAnirudh B. Vangara
19ebecbc4d
Upgrades SpectralQuant from uniform bit allocation to greedy water-filling inside the semantic subspace, with per-dimension Lloyd-Max codebooks fit to the empirical scalar marginal. Adds full three-way evaluation harness (perplexity, generation, latency, LongBench) on Modal H200 GPUs, comprehensive documentation, and the consolidated arxiv-ready manuscript. v1 is preserved on the 'v1' branch. Co-Authored-By: Anirudh B. Vangara <anirudh@sentra.app>
21 lines
728 B
Bash
21 lines
728 B
Bash
# SpectralQuant v2 — example environment variables
|
|
#
|
|
# Copy this file to `.env` and fill in real values locally. The real `.env`
|
|
# is ignored by git (.gitignore). NEVER commit a populated `.env`.
|
|
#
|
|
# Real token values must NEVER appear in any committed file.
|
|
|
|
# Hugging Face access token. The account behind this token must have
|
|
# accepted the license for `mistralai/Mistral-7B-v0.3`.
|
|
HF_TOKEN=
|
|
# Equivalent name some HF tools read; set to the same value as HF_TOKEN if needed.
|
|
HUGGING_FACE_HUB_TOKEN=
|
|
|
|
# Modal CLI tokens. Prefer `~/.modal.toml` (created by `modal token new`)
|
|
# over env vars; these are listed for completeness only.
|
|
MODAL_TOKEN_ID=
|
|
MODAL_TOKEN_SECRET=
|
|
|
|
# Optional: Weights & Biases logging.
|
|
# WANDB_API_KEY=
|