lalamo 0.5.2__py3-none-any.whl → 0.5.4__py3-none-any.whl

lalamo/speculator/inference.py

@@ -1,13 +1,15 @@
+import functools
 from collections.abc import Callable, Iterable
 from itertools import batched, chain
 from typing import NamedTuple
 
+import jax
 import jax.numpy as jnp
 
 from lalamo.data.lalamo_completions import LalamoCompletion
 from lalamo.data.utils import get_prefixes_ending_in_user_message
-from lalamo.language_model import LanguageModel
 from lalamo.message_processor import Message
+from lalamo.models import LanguageModel
 
 
 class CollectTracesEvent(NamedTuple):
@@ -25,6 +27,21 @@ def inference_collect_traces(
     tokens_to_generate: int | None = None,
     progress_callback: Callable[[CollectTracesEvent], None] | None = None,
 ) -> Iterable[LalamoCompletion]:
+    generate_tokens_compiled = (
+        jax.jit(
+            functools.partial(
+                model.generate_tokens,
+                max_output_length=max_output_length,
+                num_top_logits_to_return=num_top_logits_to_collect,
+            ),
+        )
+        .lower(
+            prompt_token_ids=jax.ShapeDtypeStruct((batch_size, max_input_length), jnp.int32),
+            prompt_lengths_without_padding=jax.ShapeDtypeStruct((batch_size,), jnp.int32),
+        )
+        .compile()
+    )
+
     prefixes = chain.from_iterable(map(get_prefixes_ending_in_user_message, conversations))
 
     tokenized_prefixes = map(model.message_processor.tokenize_request, prefixes)
@@ -32,22 +49,24 @@ def inference_collect_traces(
 
     tokens_generated, sequences_processed = 0, 0
 
-    for batch in batched(filtered_prefixes, n=batch_size):
+    for real_batch in batched(filtered_prefixes, n=batch_size):
+        batch_padding = batch_size - len(real_batch)
+        batch = (*real_batch, *(([0],) * batch_padding))
+
         length_without_padding = jnp.array(list(map(len, batch)))
-        max_len = max(map(len, batch))
 
         padded = jnp.array(
-            [jnp.pad(jnp.array(tokens), (0, max_len - len(tokens)), constant_values=0) for tokens in batch],
+            [jnp.pad(jnp.array(tokens), (0, max_input_length - len(tokens)), constant_values=0) for tokens in batch],
         )
 
-        generated = model.generate_tokens(
-            padded,
+        generated = generate_tokens_compiled(
+            prompt_token_ids=padded,
             prompt_lengths_without_padding=length_without_padding,
-            max_output_length=max_output_length,
-            num_top_logits_to_return=num_top_logits_to_collect,
         )
+
         assert generated.top_k_token_ids is not None and generated.top_k_token_logits is not None
-        for conv_idx in range(batch_size):
+
+        for conv_idx in range(len(real_batch)):
             token_ids = generated.token_ids[conv_idx].tolist()
             seqlen = next((i + 1 for i, t in enumerate(token_ids) if t in model.stop_token_ids), len(token_ids))
             if tokens_to_generate is not None:

lalamo/speculator/ngram.py

@@ -23,12 +23,14 @@ def seqhash(tokens: Iterable[int], size: int) -> int:
 
     return xxhash.xxh3_64_intdigest(packed) % size
 
-def padded_sliding_window(seq: Iterable[int], size:int, pad:int) -> Iterable[tuple[int, ...]]:
+
+def padded_sliding_window(seq: Iterable[int], size: int, pad: int) -> Iterable[tuple[int, ...]]:
     seqs = tee(seq, size)
     pads = tuple(repeat(pad, size - i) for i in range(size))
     padded_seqs = tuple(chain(pad, seq) for pad, seq in zip(pads, seqs, strict=True))
     return zip(*padded_seqs, strict=False)
 
+
 def softmax(logits: Iterable[float]) -> list[float]:
     logits = list(logits)
     log_max = max(logits)
@@ -36,9 +38,11 @@ def softmax(logits: Iterable[float]) -> list[float]:
     exp_log_sum = sum(exp_logs)
     return [exp_log / exp_log_sum for exp_log in exp_logs]
 
+
 def online_mean(old_mean: float, sample: float, new_count: int) -> float:
     return old_mean + (sample - old_mean) / new_count
 
+
 def update_probs(old_mean: dict[int, float], sample: dict[int, float], new_count: int, top_k: int) -> dict[int, float]:
     all_keys = set(old_mean.keys()).union(sample.keys())
 
@@ -115,7 +119,7 @@ class NGramSpeculator(Speculator):
         seq = list(seq)
         ngram_ctx = self.ngram_n - 1
         if ngram_ctx > 0:
-            padded_seq = [*repeat(self.ngram_pad, max(ngram_ctx - len(seq), 0)), *seq[-ngram_ctx :]]
+            padded_seq = [*repeat(self.ngram_pad, max(ngram_ctx - len(seq), 0)), *seq[-ngram_ctx:]]
         else:
             padded_seq = []
 
@@ -125,7 +129,7 @@ class NGramSpeculator(Speculator):
 
         return (
             memoryview(self.ngram_keys)[idx_start:idx_end],
-            memoryview(self.ngram_values)[idx_start:idx_end], # type: ignore (typechecker bug)
+            memoryview(self.ngram_values)[idx_start:idx_end],  # type: ignore (typechecker bug)
             memoryview(self.ngram_counts)[seq_hash : (seq_hash + 1)],
         )
 

lalamo/speculator/utils.py

@@ -19,7 +19,7 @@ def train_speculator(
 ) -> None:
     trained_tokens = 0
 
-    for trained_sequences,trace in enumerate(traces, start=1):
+    for trained_sequences, trace in enumerate(traces, start=1):
         if tokens_to_train is not None and trained_tokens + len(trace.completion_token_ids) > tokens_to_train:
             end = tokens_to_train - trained_tokens
         else:
@@ -39,7 +39,9 @@ def train_speculator(
 
 
 def test_speculator(
-    speculator: Speculator, sequence: Iterable[int] = [], max_completion_length: int = 32,
+    speculator: Speculator,
+    sequence: Iterable[int] = [],
+    max_completion_length: int = 32,
 ) -> list[int]:
     sequence = list(sequence)
     for _ in range(max_completion_length):

{lalamo-0.5.2.dist-info → lalamo-0.5.4.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lalamo
-Version: 0.5.2
+Version: 0.5.4
 Summary: JAX library for optimization and export of models for use with the UZU inference engine.
 Requires-Python: <4,>=3.12
 Description-Content-Type: text/markdown

lalamo-0.5.4.dist-info/RECORD

@@ -0,0 +1,88 @@
+lalamo/__init__.py,sha256=oyZn_-LJWGgHZqFzHw210uI-wgVJ5gjqg4yMagRXsHw,827
+lalamo/common.py,sha256=5NUFD26yQgOnEEk3LaQnce8n-VwJxILkEpFesHZhtQU,3820
+lalamo/main.py,sha256=Lqv-lU6hRSqbJeVOheZoKllK1LaPHTuR_8jNTPC7UZg,23956
+lalamo/message_processor.py,sha256=bSUAQg7CemLTnBV4LtPxJBicAalruDCA-JXjkTYPZ8U,5797
+lalamo/quantization.py,sha256=8o6ryIZLzzDYQuvBTboPfaVVdfijAKGpTxOcg3GKVD8,2752
+lalamo/registry_abc.py,sha256=ENjXiD_wEH100fNjG-W5Em1L_EQ0Lf0pdRhRGvf3qZk,2197
+lalamo/sampling.py,sha256=g_dNiJyZrRqoQIiLid4cr6nRT9N5tSz3GtHr8Bt4n-E,3404
+lalamo/utils.py,sha256=9kg5P19eaqGrSyAiNSbdfOwrv4s1PJZTHYdiNctlBSY,4368
+lalamo/data/__init__.py,sha256=exfhBLxHrg7BWutM0tAln5QuIWlNQmOhaG2noFYxfPI,189
+lalamo/data/huggingface_message.py,sha256=-7lN9eIcETQzt1Pnx3d4d8p3_I7WYMNf4mp1P91N7fI,1115
+lalamo/data/lalamo_completions.py,sha256=U_m3UNSJASUFz3rJq_taZOtL_U4B8Oj-ndkTF-JH-v4,1509
+lalamo/data/utils.py,sha256=B96gLaULyStKYuR8wjFdTpFc6YIDC8EEvGh1eiMe_Ec,338
+lalamo/model_import/__init__.py,sha256=Z8pS9rbKKx1QgUy7KZtHxiNWlZhII3mdovT9d37vAxg,168
+lalamo/model_import/common.py,sha256=dS8HbDCSReaffJ0WWdVpYjk-VGv_Ga8C3GN7z6bdxlY,11992
+lalamo/model_import/huggingface_generation_config.py,sha256=mot6VQ6ezCtEhN6VjhnvaU-nR5P5T2BuBUgpFNnWJxU,1495
+lalamo/model_import/huggingface_tokenizer_config.py,sha256=xvwdmio7b9nhn2H3uMBVligiYj58JaCFCvHY3-8dBvM,2502
+lalamo/model_import/decoder_configs/__init__.py,sha256=1ZqMcEHvCJjMIZ9iNyY31XMXOaFxB-NbqIU01BtmcEk,641
+lalamo/model_import/decoder_configs/common.py,sha256=L8PCgF5fIt3RqPlmLiJpBzDguKk9iTjk4XSItxwVG4c,3260
+lalamo/model_import/decoder_configs/executorch.py,sha256=fTEG_j-7d8riR3Fu_H5tHDjOTrWevfyw7QbWF1mUdOQ,5924
+lalamo/model_import/decoder_configs/huggingface/__init__.py,sha256=3H7GPTFNNahEvI8D1SGg2mGBgPhsIdZ213MglwbGDlE,645
+lalamo/model_import/decoder_configs/huggingface/common.py,sha256=YYIDEQy8x7lqL2qtxUHrNqfjZEiizBZ_26sTqOzjRtQ,3792
+lalamo/model_import/decoder_configs/huggingface/gemma2.py,sha256=g8LH_GlSNyL04WWi596zI0rWsD3ahnfNjDk-9zZNcDE,4759
+lalamo/model_import/decoder_configs/huggingface/gemma3.py,sha256=KlhL7y6lW_cUgsT2JjvlQbsuKZggI8DG5wazZZBk0zM,7415
+lalamo/model_import/decoder_configs/huggingface/gpt_oss.py,sha256=MBCoPbuWyzbJiBRtHOtpaPHJjQ1UVCAYcVrfIejTnlQ,7446
+lalamo/model_import/decoder_configs/huggingface/llama.py,sha256=UPeQiz2Dix8YaZYRxn9z44OZJ6c4xBQmcUZcM0Ymvh4,6934
+lalamo/model_import/decoder_configs/huggingface/llamba.py,sha256=ANB-vQK8U-zVFubZSTDXXt2S70T5SVOGzf7eOVvPzIQ,5773
+lalamo/model_import/decoder_configs/huggingface/mistral.py,sha256=MDGC0ivzJuUpOC11n8vFdcVzqccUyaRw_hkL74mVlAg,4599
+lalamo/model_import/decoder_configs/huggingface/modern_bert.py,sha256=A8nNIMhPVumvPWIFR3RexRc6XkFyUd_3mmNpmvyPEGE,8816
+lalamo/model_import/decoder_configs/huggingface/qwen2.py,sha256=n3qIANMPbtQsTtk5QEWWFZ6R85eDxR_kaZd0NDlJ3T4,5786
+lalamo/model_import/decoder_configs/huggingface/qwen3.py,sha256=lySVO-TvusAYUjDnMz-EshZHNrlNl08Xr-XqErt7K5E,6966
+lalamo/model_import/loaders/__init__.py,sha256=3THc1wQ4EPBzQkL_4EaKCa7Ev5Z7oczcvc4AHy9v5EI,228
+lalamo/model_import/loaders/common.py,sha256=kkugV-bMQlN1zvGHoj3uc7z0FbXKoMtXEBTvyu4KxK4,1844
+lalamo/model_import/loaders/executorch.py,sha256=t2Ey_mBMNC8bTSTdYWjuGXdPTRoohFlYrqtWyNkBU_8,9219
+lalamo/model_import/loaders/huggingface.py,sha256=ITA0Y_kCDFL4Tanuvd1NWUvV77WEn0VEzkcX5Whlwys,29835
+lalamo/model_import/loaders/utils.py,sha256=eiX3WKFRrAfBY-dugodscNInl5o5w3KmVcgma4atpGY,2456
+lalamo/model_import/model_specs/__init__.py,sha256=STy1Au_mwJYcl_zSei6FEp8mmGZg-Fia1tpUVAO9x9Y,1167
+lalamo/model_import/model_specs/common.py,sha256=2E6mtpUS613jsp_SDiROOvvWFB6T5U2tDkbIBXVWjmI,5780
+lalamo/model_import/model_specs/deepseek.py,sha256=Umef93_ZBuq93yYsejIRNwj3udoln1gHfrv3SK5jyMo,417
+lalamo/model_import/model_specs/gemma.py,sha256=irWgylL-pc7y3Gn5DK3fjKoCT9kJWH3B7mTa-1Gmxqc,1306
+lalamo/model_import/model_specs/gpt_oss.py,sha256=PLo0QGrXKdX61ReTRdyOaP_EH3Dmj5lp3fpJjZRwRVA,542
+lalamo/model_import/model_specs/huggingface.py,sha256=eF8ItF5reFrFkjYxwiAJcFwUAlN6CpXfM-aQ8a92ItM,430
+lalamo/model_import/model_specs/llama.py,sha256=Ml-xvRGlXBT9NJhmEpwgNo6C84oBSMYgA1_PrCYGcAw,990
+lalamo/model_import/model_specs/llamba.py,sha256=Ic3sWTv34FLJ4fG6OR_Mc5goGJQR6fa5b2WbVXbn9FA,1471
+lalamo/model_import/model_specs/mirai.py,sha256=20sLTZZQ_kn8ULfID71xdnVhjW7OjcJDf96BvVnCBUQ,605
+lalamo/model_import/model_specs/mistral.py,sha256=HAojorjOqsJn2DoMBzYRw8A70qCslhFEsE9AF5xumlg,1278
+lalamo/model_import/model_specs/pleias.py,sha256=5sRpZGYwLdsav6bLiW-459y1Cs9iJKgKkBIuGsOxtsQ,368
+lalamo/model_import/model_specs/polaris.py,sha256=Mw1-6bByjDmPIKlIUIV46CsmV5xUp_laI5Qquo5DmAQ,520
+lalamo/model_import/model_specs/qwen.py,sha256=qzLmTveATmnwNFQSFJlffcXw7syFnrCmKf9ggkkkw1Y,7050
+lalamo/model_import/model_specs/reka.py,sha256=dOUYbEMMvovQdzQuBO_DCsjGI39syhoKCvnxLkNEDCw,423
+lalamo/models/__init__.py,sha256=HfAYc4mteKu7BMyIP8aTFliLKGWJzLk1kPcdKurb8eo,243
+lalamo/models/common.py,sha256=b2woUDgHhCUEowo6tEjQb4J7iRVQvrptyRQTYMKrLis,2952
+lalamo/models/language_model.py,sha256=QPeVEyhutSze7fSNhvOvwSoYt24QMk-dtTJkos38amY,13465
+lalamo/models/router.py,sha256=7KZqHVhr2TA7Qh76KfwrvyfztfZnV-P-Ss11O8dzbRg,2013
+lalamo/modules/__init__.py,sha256=xWJ4OPAF4gKd0evYwXIK5kTnbH6nI55oLAePcoDDHQ0,3730
+lalamo/modules/activations.py,sha256=U3qTQtZawPAUcoqbkIJnmTYcaNiQuSPMLcBeJ398GhI,1022
+lalamo/modules/classifier.py,sha256=_jtJ3INEq1dJP5HpUmcDk9YYzpRYlQ04zvFGaWBV6Lg,12101
+lalamo/modules/common.py,sha256=dqDEOi-C3H4U9iWUisU32RA-wRDCGuaUNGbObRBhyQM,3315
+lalamo/modules/decoder.py,sha256=Opd3QIq1mpGr9P7sLH-Fryitlfp6ESTpcX71vgm89t0,7129
+lalamo/modules/embedding.py,sha256=yhVZLg-yb1RwEXoq8pZx-08MPHLEOraPv73gH8shQ_o,21782
+lalamo/modules/linear.py,sha256=XfIYhmpk-bwNHIzIgsL48ZUTclHD2KB4uXHMw9NTE-8,42991
+lalamo/modules/mlp.py,sha256=bL3sQ46vCNt1MBRwlzmXZx9nQfRe4axpGe5UOFVanBI,17959
+lalamo/modules/mlx_interop.py,sha256=FdfU_1iES-HQ9r4K0SkYwJTyvE0f-_T5ursNCjPLZKY,467
+lalamo/modules/normalization.py,sha256=cBdOq6OmJssunVeEwFRJD0BDhgFAN7J8gOKwzIUAY8I,3005
+lalamo/modules/rope.py,sha256=rCik7vBNqRXYg3LGbmc1mezPRNbIYMg5cydTFpQy-eU,10157
+lalamo/modules/torch_interop.py,sha256=-mujd1zI4ec2w92Hd50RtDa0K3jl6ZSnPxc5r3Fp9nU,916
+lalamo/modules/transformer.py,sha256=67-WZX2eE314abiQOhRNSooTHeJh4q9mlZQIxQ-oASU,10162
+lalamo/modules/transformer_layer.py,sha256=CfkYIn8a3pR4PPsI9hmAXpyiTbjXo-Gzl2OU9lAQlkI,12724
+lalamo/modules/utils.py,sha256=t_TayWT6g5LtYKhJaod-u_COWaI_VbNd3eYek9Nj0lc,441
+lalamo/modules/token_mixers/__init__.py,sha256=_t4T25C4WBVJQ1SqkQPGrrAc7bPKhDO3K2btgefVh5s,909
+lalamo/modules/token_mixers/attention.py,sha256=gkGMFah2OHB_tyJpkshM1KhMnzG6U7Xt273MkBvDk58,16584
+lalamo/modules/token_mixers/common.py,sha256=-ej1pIrrp845ztavJ3oh82U3HEsV_rEHxMZOEDp7iK8,1979
+lalamo/modules/token_mixers/mamba.py,sha256=MIIMZAlVVE4YwyT0PsxA0OWXa13ondoJchRxQbhq678,18797
+lalamo/modules/token_mixers/state/__init__.py,sha256=iQaX7njz3XtwGugI5_PUOIp1wdCzd5h08UkgF6yW3zo,307
+lalamo/modules/token_mixers/state/common.py,sha256=dcwBevAdeJpBjf7_YRk7TKrJHsCnpljhfzZy-3h9898,661
+lalamo/modules/token_mixers/state/kv_cache.py,sha256=QfnS3XgSmyDI9MBUbeLI4ABHLxiMcXDbZsqe0fd3KQo,8788
+lalamo/modules/token_mixers/state/mamba_state.py,sha256=LHzJvNE6MkB7nrsZSNto6pxbnMJCl--JOoe9Fkcc9Mg,1642
+lalamo/speculator/__init__.py,sha256=9-tmZcbCom_lIGpJYn6xLlnEahFLFidpqmgkafmu--k,456
+lalamo/speculator/common.py,sha256=PudF_gkpe5_nQ-57sAC-foE1xCy_H2Axh5KwRoA86lo,587
+lalamo/speculator/estimator.py,sha256=SFF24TtxGVbb3hEziZ40A7aNdBd6_ZZJSX-CqmPhvA8,2503
+lalamo/speculator/inference.py,sha256=EzNLmdP-AOtIN8C39-fq-3ZnKX_yxOOwMufDEmi54zM,3624
+lalamo/speculator/ngram.py,sha256=95mdfAWhx4d5XOnOwhyhElnvcy6nlUjYhcbJzqDs414,5875
+lalamo/speculator/utils.py,sha256=0wZoMMIzzk0Q-3zq5H5f-JBplePNHxywndkrNtOJOyo,1697
+lalamo-0.5.4.dist-info/licenses/LICENSE,sha256=diHRfjSEJHD1nnEeMIfMRCjR3UERf8bT3eseD6b1ayA,1072
+lalamo-0.5.4.dist-info/METADATA,sha256=-OqjgnggM7Q7RMOcH7Dsin68vVxbJGfwW2LxUslz-ro,3146
+lalamo-0.5.4.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+lalamo-0.5.4.dist-info/entry_points.txt,sha256=qli7qTfnBk5WP10rOGXXEckHMtt-atJMDWd8jN89Uks,43
+lalamo-0.5.4.dist-info/top_level.txt,sha256=VHvWL5JN5XRG36NsN_MieJ7EwRihEOrEjyDaTdFJ-aI,7
+lalamo-0.5.4.dist-info/RECORD,,

lalamo-0.5.2.dist-info/RECORD

@@ -1,80 +0,0 @@
-lalamo/__init__.py,sha256=3P0i5yWyx-A6xFJNXobXqcLYXbgpLKPr4tL-x3OAJto,487
-lalamo/common.py,sha256=5NUFD26yQgOnEEk3LaQnce8n-VwJxILkEpFesHZhtQU,3820
-lalamo/language_model.py,sha256=ZnNwqxT8yCfIE0TSaF8Vhva8GRHzDzEoUb95VjAfOtM,14510
-lalamo/main.py,sha256=YNO7s0H04Jg5OYmKEGmdIJLzW1W_-uJj0TxJY_Tsc0M,19028
-lalamo/message_processor.py,sha256=eDKqOGRTAc4rbwzye0FAGwTnG_nIlerdlLEcn8QlT0Q,5817
-lalamo/quantization.py,sha256=8o6ryIZLzzDYQuvBTboPfaVVdfijAKGpTxOcg3GKVD8,2752
-lalamo/registry_abc.py,sha256=ENjXiD_wEH100fNjG-W5Em1L_EQ0Lf0pdRhRGvf3qZk,2197
-lalamo/sampling.py,sha256=g_dNiJyZrRqoQIiLid4cr6nRT9N5tSz3GtHr8Bt4n-E,3404
-lalamo/utils.py,sha256=9kg5P19eaqGrSyAiNSbdfOwrv4s1PJZTHYdiNctlBSY,4368
-lalamo/data/__init__.py,sha256=9-lii4yjDZkZFkSNJA0Wd5ZOcxnhs5390IWO4JrMMWI,190
-lalamo/data/huggingface_message.py,sha256=AJi4t1DIJ0AtFPQW8rBxNwjnve8RLl5WuZH9DFVxRqU,1114
-lalamo/data/lalamo_completions.py,sha256=U_m3UNSJASUFz3rJq_taZOtL_U4B8Oj-ndkTF-JH-v4,1509
-lalamo/data/utils.py,sha256=B96gLaULyStKYuR8wjFdTpFc6YIDC8EEvGh1eiMe_Ec,338
-lalamo/model_import/__init__.py,sha256=Z8pS9rbKKx1QgUy7KZtHxiNWlZhII3mdovT9d37vAxg,168
-lalamo/model_import/common.py,sha256=EW3JwXGfjoUj98DEmoNAUJn1lHRJ8yXpF_D38efrxv8,8782
-lalamo/model_import/huggingface_generation_config.py,sha256=mot6VQ6ezCtEhN6VjhnvaU-nR5P5T2BuBUgpFNnWJxU,1495
-lalamo/model_import/huggingface_tokenizer_config.py,sha256=xvwdmio7b9nhn2H3uMBVligiYj58JaCFCvHY3-8dBvM,2502
-lalamo/model_import/decoder_configs/__init__.py,sha256=lv00ejNtoUtMVQjpjbNiL8yZCIalyutYPfIOKyg5kfU,545
-lalamo/model_import/decoder_configs/common.py,sha256=9dW704HdLan_E3JBT9wQftuU5TC-shGC0q6SGr6iljY,1851
-lalamo/model_import/decoder_configs/executorch.py,sha256=tyJ-oP8wezOHBGL7Y9Iy3nplHprUHSo7pkzrvEPk_7k,5725
-lalamo/model_import/decoder_configs/huggingface/__init__.py,sha256=m17Pz3pw7lHXrcw3FGE15TNUu-Qk7m9CYQSXfS3CWD0,575
-lalamo/model_import/decoder_configs/huggingface/common.py,sha256=BJekEs9iStsygVa9QcPItok5AQIq8PvMZf7xAmykl-w,3043
-lalamo/model_import/decoder_configs/huggingface/gemma2.py,sha256=0pO4IinzqB7hLoEUMlVnJfN3ncerTgZKIrMZPs3soR8,4562
-lalamo/model_import/decoder_configs/huggingface/gemma3.py,sha256=r8Nfy-9EPXhIkhJkvx0P-7NFLMaku_LJIEhx3HziShk,7217
-lalamo/model_import/decoder_configs/huggingface/gpt_oss.py,sha256=8TWptvJzVJtEtyWrkKaJA6LLORSKLolxv3gwLtDC0uA,7248
-lalamo/model_import/decoder_configs/huggingface/llama.py,sha256=amMhyN_QEscFmOvjmghcBg9VXBDqE-mpp0Y5Gj0tSp0,6735
-lalamo/model_import/decoder_configs/huggingface/llamba.py,sha256=dJVD_84suhF5sQLRvlsfMDmrrTNhIR6AdMGOm54rC6U,5482
-lalamo/model_import/decoder_configs/huggingface/mistral.py,sha256=EjDIoGhxyIjugoRFlEIM8QZkxbGiaIeyTTQx8WZEkWw,4460
-lalamo/model_import/decoder_configs/huggingface/qwen2.py,sha256=mV_5IkdYEMJQnrTpfXTc_sJzWx0dSXKawuK5boWnYJE,5587
-lalamo/model_import/decoder_configs/huggingface/qwen3.py,sha256=S4LIypTvfmZ6d862MAK2gZ3kWDCCpQoJ8tLrIhbXbPo,6769
-lalamo/model_import/loaders/__init__.py,sha256=Olg7a79phusilNgEa7PTgx1JgQQJLgAVg18T8isp0mw,148
-lalamo/model_import/loaders/common.py,sha256=kkugV-bMQlN1zvGHoj3uc7z0FbXKoMtXEBTvyu4KxK4,1844
-lalamo/model_import/loaders/executorch.py,sha256=NmwuB4MvTO_n3iNCd6jNobRuFOf57d1uMiUEkEaihM8,9002
-lalamo/model_import/loaders/huggingface.py,sha256=IGWlVErAoEyIhS8qcoiPP8TTJmHmyO2enFOhthzSByg,22779
-lalamo/model_import/loaders/utils.py,sha256=eiX3WKFRrAfBY-dugodscNInl5o5w3KmVcgma4atpGY,2456
-lalamo/model_import/model_specs/__init__.py,sha256=gdWPNkjvs2jZ9WwGs81tjLNx1WIOW3E5XuWEOHkI0f4,1076
-lalamo/model_import/model_specs/common.py,sha256=ZNK46wgmotI9XdoEe04D_hq-ElijJLm2QhVSBbE24lg,5602
-lalamo/model_import/model_specs/deepseek.py,sha256=Umef93_ZBuq93yYsejIRNwj3udoln1gHfrv3SK5jyMo,417
-lalamo/model_import/model_specs/gemma.py,sha256=YGWM-J7jBEL16c_LZ9F_6dgeZnf4sEPLDKngl_3FbrE,1289
-lalamo/model_import/model_specs/gpt_oss.py,sha256=PLo0QGrXKdX61ReTRdyOaP_EH3Dmj5lp3fpJjZRwRVA,542
-lalamo/model_import/model_specs/huggingface.py,sha256=TEkU8y95_hmUWyF-Q5hn0dE2SvXbApghAsQwhWRu4D0,431
-lalamo/model_import/model_specs/llama.py,sha256=Ml-xvRGlXBT9NJhmEpwgNo6C84oBSMYgA1_PrCYGcAw,990
-lalamo/model_import/model_specs/llamba.py,sha256=Ic3sWTv34FLJ4fG6OR_Mc5goGJQR6fa5b2WbVXbn9FA,1471
-lalamo/model_import/model_specs/mistral.py,sha256=HAojorjOqsJn2DoMBzYRw8A70qCslhFEsE9AF5xumlg,1278
-lalamo/model_import/model_specs/pleias.py,sha256=5sRpZGYwLdsav6bLiW-459y1Cs9iJKgKkBIuGsOxtsQ,368
-lalamo/model_import/model_specs/polaris.py,sha256=Mw1-6bByjDmPIKlIUIV46CsmV5xUp_laI5Qquo5DmAQ,520
-lalamo/model_import/model_specs/qwen.py,sha256=qzLmTveATmnwNFQSFJlffcXw7syFnrCmKf9ggkkkw1Y,7050
-lalamo/model_import/model_specs/reka.py,sha256=dOUYbEMMvovQdzQuBO_DCsjGI39syhoKCvnxLkNEDCw,423
-lalamo/modules/__init__.py,sha256=dok0jZSDknlZ5uDVUckMLFGtFkLtgfm1Msy7aLNFx1Y,3421
-lalamo/modules/activations.py,sha256=U3qTQtZawPAUcoqbkIJnmTYcaNiQuSPMLcBeJ398GhI,1022
-lalamo/modules/common.py,sha256=f7gRmnEP4NMjoTuoqwaBaSs7pf3LFAYEE5oojbM1mKI,3214
-lalamo/modules/decoder.py,sha256=IdTRmD47uSSK1pPYBhd4l_cOZQNzPLichHytUxVloEw,12274
-lalamo/modules/decoder_layer.py,sha256=oHBzWR7m0-aP4SMgjwF7cJPdaJInNzFY_kWiAl1yGE8,11991
-lalamo/modules/embedding.py,sha256=yhVZLg-yb1RwEXoq8pZx-08MPHLEOraPv73gH8shQ_o,21782
-lalamo/modules/linear.py,sha256=XfIYhmpk-bwNHIzIgsL48ZUTclHD2KB4uXHMw9NTE-8,42991
-lalamo/modules/mlp.py,sha256=9_4OpUw4hDFT9X2jbakf8lm5ZvHhQVwHThArH02AySw,17704
-lalamo/modules/mlx_interop.py,sha256=FdfU_1iES-HQ9r4K0SkYwJTyvE0f-_T5ursNCjPLZKY,467
-lalamo/modules/normalization.py,sha256=gdmv4tWLk8CwvIfoS0rY3QTVRXhjHQNNF6YGYjyez1c,2790
-lalamo/modules/rope.py,sha256=rCik7vBNqRXYg3LGbmc1mezPRNbIYMg5cydTFpQy-eU,10157
-lalamo/modules/torch_interop.py,sha256=-mujd1zI4ec2w92Hd50RtDa0K3jl6ZSnPxc5r3Fp9nU,916
-lalamo/modules/utils.py,sha256=t_TayWT6g5LtYKhJaod-u_COWaI_VbNd3eYek9Nj0lc,441
-lalamo/modules/token_mixers/__init__.py,sha256=_t4T25C4WBVJQ1SqkQPGrrAc7bPKhDO3K2btgefVh5s,909
-lalamo/modules/token_mixers/attention.py,sha256=2NtruCEMdTuk2H-cD9uEhF18pwWmQyJa0mOz0RnoSi4,16501
-lalamo/modules/token_mixers/common.py,sha256=-ej1pIrrp845ztavJ3oh82U3HEsV_rEHxMZOEDp7iK8,1979
-lalamo/modules/token_mixers/mamba.py,sha256=MIIMZAlVVE4YwyT0PsxA0OWXa13ondoJchRxQbhq678,18797
-lalamo/modules/token_mixers/state/__init__.py,sha256=iQaX7njz3XtwGugI5_PUOIp1wdCzd5h08UkgF6yW3zo,307
-lalamo/modules/token_mixers/state/common.py,sha256=dcwBevAdeJpBjf7_YRk7TKrJHsCnpljhfzZy-3h9898,661
-lalamo/modules/token_mixers/state/kv_cache.py,sha256=KgMwZC_RmBn_Cw-85gILtEjUVP7sNpXPkwyWz37RWak,8536
-lalamo/modules/token_mixers/state/mamba_state.py,sha256=LHzJvNE6MkB7nrsZSNto6pxbnMJCl--JOoe9Fkcc9Mg,1642
-lalamo/speculator/__init__.py,sha256=pLGwZiXZZeH2SaMsIGiqHzXs5AJtbxWweXaKc4fPFHc,262
-lalamo/speculator/common.py,sha256=PudF_gkpe5_nQ-57sAC-foE1xCy_H2Axh5KwRoA86lo,587
-lalamo/speculator/inference.py,sha256=4E7_8jwLwg1bvUUH59cLxjD73-IJ9xv58yCwYMFa1r4,3101
-lalamo/speculator/ngram.py,sha256=3O9akE2IGN_7Mp6l_zwhUpdNvjZB0rAlQm51EuX_-rY,5869
-lalamo/speculator/utils.py,sha256=K5qDGEdWYD47C1xLJQ3bamIHAG3-VEuANf-821VeanE,1688
-lalamo-0.5.2.dist-info/licenses/LICENSE,sha256=diHRfjSEJHD1nnEeMIfMRCjR3UERf8bT3eseD6b1ayA,1072
-lalamo-0.5.2.dist-info/METADATA,sha256=bEhxBxyO5vmbF5KyrEMRSATIzJ55M9HfunCkllcC-VY,3146
-lalamo-0.5.2.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-lalamo-0.5.2.dist-info/entry_points.txt,sha256=qli7qTfnBk5WP10rOGXXEckHMtt-atJMDWd8jN89Uks,43
-lalamo-0.5.2.dist-info/top_level.txt,sha256=VHvWL5JN5XRG36NsN_MieJ7EwRihEOrEjyDaTdFJ-aI,7
-lalamo-0.5.2.dist-info/RECORD,,