github/tinygrad
1
1
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-04-29 03:00:14 -04:00
Code Issues Packages Projects Releases Wiki Activity

llm is the only app (#15779)

Browse Source 
* tinygrad/llm is the only app

* upd pyproject

* claude refs

* scoping

* min diff
This commit is contained in:
George Hotz
2026-04-17 10:44:48 +08:00
committed by GitHub
parent 0e69388f6b
commit ec00cefa5b
16 changed files with 49 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
test/unit/test_attention.py
View File

@@ -1,7 +1,7 @@
import unittest
import numpy as np
from tinygrad import Tensor, dtypes
from tinygrad.apps.llm import (
from tinygrad.llm.cli import (
GatedDeltaNetBlock, SSMConfig, TransformerBlock, TransformerConfig,
apply_rope as apply_rope_new, precompute_freqs_cis, pairwise_topk,
)

4
test/unit/test_llm_mla.py
View File

@@ -1,7 +1,7 @@
import unittest
import numpy as np
from tinygrad import Tensor
from tinygrad.apps.llm import Transformer, TransformerConfig, apply_rope
from tinygrad.llm.cli import Transformer, TransformerConfig, apply_rope
class TestMLA(unittest.TestCase):
def _make_config(self, **kwargs):
@@ -13,7 +13,7 @@ class TestMLA(unittest.TestCase):
def test_mla_attention_matches_naive(self):
config = self._make_config(max_context=16)
from tinygrad.apps.llm import MLATransformerBlock, precompute_freqs_cis
from tinygrad.llm.cli import MLATransformerBlock, precompute_freqs_cis
block = MLATransformerBlock(config)
c = config

2
test/unit/test_llm_moe.py
View File

@@ -2,7 +2,7 @@ import unittest
import numpy as np
from dataclasses import replace
from tinygrad import Tensor
from tinygrad.apps.llm import TransformerBlock, TransformerConfig
from tinygrad.llm.cli import TransformerBlock, TransformerConfig
def _moe_config(dim=8, hidden=16, n_heads=2, num_experts=4, num_experts_per_tok=2):
return TransformerConfig(

2
test/unit/test_llm_server.py
View File

@@ -2,7 +2,7 @@ import unittest
from unittest.mock import patch
from tinygrad import Tensor, UOp
from tinygrad.schedule import schedule_cache
from tinygrad.apps.llm import Transformer, TransformerConfig
from tinygrad.llm.cli import Transformer, TransformerConfig
TEST_CONFIG = TransformerConfig(num_blocks=1, dim=64, hidden_dim=128, n_heads=2, n_kv_heads=2,
norm_eps=1e-5, vocab_size=100, head_dim=32, rope_theta=10000.0, rope_dim=32, v_head_dim=32, max_context=32)