Skip to content

setup: count the 3-bit models' 262K context against RAM, not a fixed 90 GB #134

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
architectds wants to merge 1 commit into from
Closed

setup: count the 3-bit models' 262K context against RAM, not a fixed 90 GB #134

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 8 additions & 3 deletions setup.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1480,9 +1480,14 @@ def main() -> int:
for i, c in enumerate(CONTEXTS, 1):
say(f" {i}) {c // 1024}K tokens" + (" (recommended for your GPU)" if c == rec_ctx else ""))
ctx = CONTEXTS[int(ask("Context?", [str(i) for i in range(1, 6)], str(CONTEXTS.index(rec_ctx) + 1), a.yes)) - 1]
if model in ("IQ3_XXS", "IQ3_S") and ram < 90 and ctx > 131072:
warn(f"{model} with a 262K context needs more than 64 GB of RAM ({MODELS[model]['arena_gb']:.0f} GB of experts "
"+ the context): using 128K")
# The experts' arena plus the context's KV (in RAM from 64K up: ~13.7 KB/token at 8 bits) must fit, with room
# for everything else. Counted, not a fixed 90 GB: 24 GB of room keeps the rule for 64 GB PCs as it was (128K
# for both 3-bit models), while a box with ~84 GB keeps the 262K it asked for (measured: a Colab A100-40G runs
# IQ3_S at 262K with images in 56 of its 83.5 GiB).
need_gb = MODELS[model].get("arena_gb", 0) + ctx * 13 * 1056 / 1e9 + 24
if model in ("IQ3_XXS", "IQ3_S") and ram < need_gb and ctx > 131072:
warn(f"{model} with a {ctx // 1024}K context needs about {need_gb:.0f} GB of RAM "
f"({MODELS[model]['arena_gb']:.0f} GB of experts + the context + room for the rest): using 128K")
ctx = 131072
ok(f"context: {ctx} tokens")
# the KV cache (the model's memory of the conversation): 8-bit, or 4-bit after a Hadamard rotation (PR #21)
Expand Down