llm is the only app (#15779)

* tinygrad/llm is the only app

* upd pyproject

* claude refs

* scoping

* min diff

test/external/external_llm_eval.py

@@ -1,4 +1,4 @@
-# eval for tinygrad.apps.llm -- hits the server via OpenAI API
+# eval for OpenAI API server
 # uses Meta's exact ARC-Challenge prompt template from lm-evaluation-harness llama3 tasks
 import argparse, re, pyarrow.parquet as pq
 from openai import OpenAI

test/external/external_test_simple_tokenizer.py

@@ -1,7 +1,7 @@
 import functools, multiprocessing
 from transformers import AutoTokenizer
 from datasets import load_dataset
-from tinygrad.apps.llm import SimpleTokenizer
+from tinygrad.llm.cli import SimpleTokenizer
 from tinygrad.helpers import tqdm, getenv, partition
 
 @functools.cache

test/null/test_attention.py

@@ -1,6 +1,6 @@
 import unittest
 from tinygrad import Tensor, dtypes, TinyJit, UOp
-from tinygrad.apps.llm import apply_rope as apply_rope_new, precompute_freqs_cis
+from tinygrad.llm.cli import apply_rope as apply_rope_new, precompute_freqs_cis
 from test.helpers import assert_jit_cache_len
 
 def apply_rope(x:Tensor, start_pos:int):

test/null/test_llm_server.py

@@ -22,18 +22,15 @@ class TestLLMServer(unittest.TestCase):
     cls.bos_id = 1
     cls.eos_id = 999
 
-    import tinygrad.apps.llm as llm_module
-    llm_module.model = cls.mock_model
-    llm_module.model_name = "test-model"
-    llm_module.tok = cls.mock_tok
-    llm_module.bos_id = cls.bos_id
-    llm_module.eos_id = cls.eos_id
-    llm_module.eot_id = None
+    from tinygrad.llm.cli import Handler, LLMServer
 
-    from tinygrad.apps.llm import Handler
-    from tinygrad.viz.serve import TCPServerWithReuse
-
-    cls.server = TCPServerWithReuse(('127.0.0.1', 0), Handler)
+    cls.server = LLMServer(('127.0.0.1', 0), Handler)
+    cls.server.model = cls.mock_model
+    cls.server.model_name = "test-model"
+    cls.server.tok = cls.mock_tok
+    cls.server.bos_id = cls.bos_id
+    cls.server.eos_id = cls.eos_id
+    cls.server.eot_id = None
     cls.port = cls.server.server_address[1]
     cls.server_thread = threading.Thread(target=cls.server.serve_forever, daemon=True)
     cls.server_thread.start()

test/null/test_llm_tokenizer.py

@@ -1,5 +1,5 @@
 import unittest, base64, functools, sys
-from tinygrad.apps.llm import SimpleTokenizer
+from tinygrad.llm.cli import SimpleTokenizer
 from tinygrad.helpers import fetch
 
 @unittest.skipIf(sys.platform == 'win32', "fetch race condition on Windows")

test/unit/test_attention.py

@@ -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,
 )

test/unit/test_llm_mla.py

@@ -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

test/unit/test_llm_moe.py

@@ -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(

test/unit/test_llm_server.py

@@ -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)