github/tinygrad
1
1
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-01-06 21:53:53 -05:00
Code Issues Packages Projects Releases Wiki Activity

replace hardcoded GPU in llama debug msg (#12102)

Browse Source
This commit is contained in:
chenyu
2025-09-10 13:56:40 -04:00
committed by GitHub
parent 5a84d86db7
commit 0599e86186
4 changed files with 6 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
examples/gpt2.py
View File

@@ -189,7 +189,7 @@ class GPT2:
GlobalCounters.reset()
if timing: print("")
st = GlobalCounters.time_sum_s
with Timing("ran model in ", on_exit=(lambda et: (f", {(GlobalCounters.time_sum_s-st)*1e3:.2f} ms on GPU" if DEBUG>=2 else "")+
with Timing("ran model in ", on_exit=(lambda et: (f", {(GlobalCounters.time_sum_s-st)*1e3:.2f} ms on {Device.DEFAULT}" if DEBUG>=2 else "")+
f", {GlobalCounters.global_ops*1e-9:.2f} GOPS, {GlobalCounters.global_mem*1e-9:.2f} GB"+
(f", {GlobalCounters.global_mem*1e-9/(GlobalCounters.time_sum_s-st):.2f} GB/s" if DEBUG>=2 else "")) if DEBUG else None, enabled=timing):
with WallTimeEvent(BenchEvent.STEP):

2
examples/llama.py
View File

@@ -478,7 +478,7 @@ After you are done speaking, output [EOS]. You are not Chad.
with Profiling(enabled=args.profile):
with Timing("total ", enabled=args.timing, on_exit=lambda x: f", {1e9/x:.2f} tok/s, {GlobalCounters.global_mem/x:.2f} GB/s, param {param_bytes/x:.2f} GB/s"):
with WallTimeEvent(BenchEvent.STEP):
with Timing("enqueue in ", on_exit=(lambda et: (f", {(GlobalCounters.time_sum_s-st)*1e3:.2f} ms on GPU" if DEBUG>=2 else "")+
with Timing("enqueue in ", on_exit=(lambda et: (f", {(GlobalCounters.time_sum_s-st)*1e3:.2f} ms on {Device.DEFAULT}" if DEBUG>=2 else "")+
f", {GlobalCounters.global_ops*1e-9:.2f} GOPS, {GlobalCounters.global_mem*1e-9:.2f} GB"+
(f", {GlobalCounters.global_mem*1e-9/(GlobalCounters.time_sum_s-st):.2f} GB/s, param {param_bytes*1e-9/(GlobalCounters.time_sum_s-st):.2f} GB/s" if DEBUG>=2 else "")) if DEBUG else None, enabled=args.timing):
tok_tensor = llama.model(next_tok, start_pos, args.temperature)

4
examples/llama3.py
View File

@@ -441,7 +441,7 @@ if __name__ == "__main__":
with Profiling(enabled=args.profile):
with Timing("total ", on_exit=lambda x: f", {1e9/x:.2f} tok/s, {GlobalCounters.global_mem/x:.2f} GB/s, param {param_bytes/x:.2f} GB/s"):
with WallTimeEvent(BenchEvent.STEP):
with Timing("enqueue in ", on_exit=(lambda et: (f", {(GlobalCounters.time_sum_s-st)*1e3:.2f} ms on GPU" if DEBUG>=2 else "")+
with Timing("enqueue in ", on_exit=(lambda et: (f", {(GlobalCounters.time_sum_s-st)*1e3:.2f} ms on {Device.DEFAULT}" if DEBUG>=2 else "")+
f", {GlobalCounters.global_ops*1e-9:.2f} GOPS, {GlobalCounters.global_mem*1e-9:.2f} GB"+
(f", {GlobalCounters.global_mem*1e-9/(GlobalCounters.time_sum_s-st):.2f} GB/s, param {param_bytes*1e-9/(GlobalCounters.time_sum_s-st):.2f} GB/s" if DEBUG>=2 else "")) if DEBUG else None):
tok = model(Tensor([[last_tok]], device=device), start_pos, TEMPERATURE, TOP_K, TOP_P, ALPHA_F, ALPHA_P)
@@ -479,7 +479,7 @@ if __name__ == "__main__":
st = GlobalCounters.time_sum_s
with Profiling(enabled=args.profile):
with Timing("total ", enabled=args.timing, on_exit=lambda x: f", {1e9/x:.2f} tok/s, {GlobalCounters.global_mem/x:.2f} GB/s, param {param_bytes/x:.2f} GB/s"):
with Timing("enqueue in ", on_exit=(lambda et: (f", {(GlobalCounters.time_sum_s-st)*1e3:.2f} ms on GPU" if DEBUG>=2 else "")+
with Timing("enqueue in ", on_exit=(lambda et: (f", {(GlobalCounters.time_sum_s-st)*1e3:.2f} ms on {Device.DEFAULT}" if DEBUG>=2 else "")+
f", {GlobalCounters.global_ops*1e-9:.2f} GOPS, {GlobalCounters.global_mem*1e-9:.2f} GB"+
(f", {GlobalCounters.global_mem*1e-9/(GlobalCounters.time_sum_s-st):.2f} GB/s, param {param_bytes*1e-9/(GlobalCounters.time_sum_s-st):.2f} GB/s" if DEBUG>=2 else "")) if DEBUG else None, enabled=args.timing):

4
examples/qwq.py
View File

@@ -8,7 +8,7 @@ from typing import Dict, Union
from extra.models.llama import Transformer, convert_from_huggingface, fix_bf16
from examples.llama3 import load
from tinygrad import nn, Tensor
from tinygrad import nn, Tensor, Device
from tinygrad.helpers import fetch, colored, GlobalCounters, Timing, DEBUG
from tinygrad.nn.state import load_state_dict, get_parameters
@@ -80,7 +80,7 @@ if __name__ == "__main__":
st = GlobalCounters.time_sum_s
next_tok = Tensor([toks[start_pos:]]) if tok_tensor is None or (len(toks)-start_pos) > 1 else tok_tensor.reshape(1, 1)
with Timing("total ", enabled=args.timing, on_exit=lambda x: f", {1e9/x:.2f} tok/s, {GlobalCounters.global_mem/x:.2f} GB/s, param {param_bytes/x:.2f} GB/s"):
with Timing("enqueue in ", on_exit=(lambda et: (f", {(GlobalCounters.time_sum_s-st)*1e3:.2f} ms on GPU" if DEBUG>=2 else "") +
with Timing("enqueue in ", on_exit=(lambda et: (f", {(GlobalCounters.time_sum_s-st)*1e3:.2f} ms on {Device.DEFAULT}" if DEBUG>=2 else "") +
f", {GlobalCounters.global_ops*1e-9:.2f} GOPS, {GlobalCounters.global_mem*1e-9:.2f} GB" +
(f", {GlobalCounters.global_mem*1e-9/(GlobalCounters.time_sum_s-st):.2f} GB/s, param {param_bytes*1e-9/(GlobalCounters.time_sum_s-st):.2f} GB/s" if DEBUG>=2 else "")) if DEBUG else None, enabled=args.timing):
tok_tensor = transformer(next_tok, start_pos, args.temperature)