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

coder.py can write and run code (#2439)

Browse Source 
* wip mistral

* coder

* touchups

* cleanups

* mistral cleanups

* clean up cache create

* download the weights, fix tests

* fix llama loading

* global fixup

* clean up all

* move llama model

* cleanups

* Revert "cleanups"

This reverts commit a71c5d59eb.

* fine, leave it
This commit is contained in:
George Hotz
2023-11-25 12:27:54 -08:00
committed by GitHub
parent df41a57e09
commit 7170a9a057
10 changed files with 334 additions and 167 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
test/external/external_test_allocator_on_models.py vendored
View File

@@ -93,7 +93,7 @@ class TestAllocators(unittest.TestCase):
old_type = Tensor.default_type
Tensor.default_type = dtypes.float16
args_tiny = {"dim": 1024, "multiple_of": 256, "n_heads": 8, "n_layers": 8, "norm_eps": 1e-05, "vocab_size": 1000}
args_tiny = {"dim": 1024, "hidden_dim": 1024, "n_heads": 8, "n_layers": 8, "norm_eps": 1e-05, "vocab_size": 1000}
def __test():
model = Transformer(**args_tiny)
derandomize_model(model)
@@ -105,7 +105,7 @@ class TestAllocators(unittest.TestCase):
@unittest.skipUnless(Device.DEFAULT == "GPU", "Not Implemented")
def test_lru_allocator_tiny_llama_alloc_counts(self):
args_tiny = {"dim": 1024, "multiple_of": 256, "n_heads": 8, "n_layers": 8, "norm_eps": 1e-05, "vocab_size": 1000}
args_tiny = {"dim": 1024, "hidden_dim": 1024, "n_heads": 8, "n_layers": 8, "norm_eps": 1e-05, "vocab_size": 1000}
def test_alloc_count(t):
model = Transformer(**args_tiny)
for v in get_state_dict(model).values(): v.assign(Tensor.empty(*v.shape, dtype=v.dtype))

2
test/external/external_test_jit_on_models.py vendored
View File

@@ -19,7 +19,7 @@ class TestJittedModels(unittest.TestCase):
old_type = Tensor.default_type
Tensor.default_type = dtypes.float16
args_tiny = {"dim": 1024, "multiple_of": 256, "n_heads": 8, "n_layers": 8, "norm_eps": 1e-05, "vocab_size": 1000}
args_tiny = {"dim": 1024, "hidden_dim": 1024, "n_heads": 8, "n_layers": 8, "norm_eps": 1e-05, "vocab_size": 1000}
model = Transformer(**args_tiny)
derandomize_model(model)
def test(t): return model(t, 0).realize()

2
test/external/external_test_opt.py vendored
View File

@@ -86,7 +86,7 @@ class TestInferenceMinKernels(unittest.TestCase):
def test_llama(self):
from examples.llama import Transformer
from tinygrad.shape.symbolic import Variable
args_tiny = {"dim": 512, "multiple_of": 256, "n_heads": 8, "n_layers": 4, "norm_eps": 1e-05, "vocab_size": 1000}
args_tiny = {"dim": 512, "hidden_dim": 1024, "n_heads": 8, "n_layers": 4, "norm_eps": 1e-05, "vocab_size": 1000}
model = Transformer(**args_tiny)
for p in get_parameters(model): p.assign(np.zeros(p.shape, dtype=p.dtype.np))
with CLCache(100):

2
test/models/test_real_world.py
View File

@@ -51,7 +51,7 @@ class TestRealWorld(unittest.TestCase):
def test_llama(self):
Tensor.default_type = dtypes.float16
args_tiny = {"dim": 1024, "multiple_of": 256, "n_heads": 8, "n_layers": 8, "norm_eps": 1e-05, "vocab_size": 1000}
args_tiny = {"dim": 1024, "hidden_dim": 2048, "n_heads": 8, "n_layers": 8, "norm_eps": 1e-05, "vocab_size": 1000}
model = LLaMaTransformer(**(args_tiny if CI else LLAMA_MODEL_PARAMS["1"]["7B"]["args"]))
derandomize_model(model)
@TinyJit