optimum-rbln 0.8.2a7__py3-none-any.whl → 0.8.2rc0__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Potentially problematic release.
This version of optimum-rbln might be problematic. Click here for more details.
- optimum/rbln/__init__.py +4 -0
- optimum/rbln/__version__.py +2 -2
- optimum/rbln/transformers/__init__.py +4 -0
- optimum/rbln/transformers/models/__init__.py +4 -0
- optimum/rbln/transformers/models/decoderonly/decoderonly_architecture.py +1 -2
- optimum/rbln/transformers/models/decoderonly/modeling_decoderonly.py +1 -1
- optimum/rbln/transformers/models/gemma3/configuration_gemma3.py +11 -0
- optimum/rbln/transformers/models/gemma3/modeling_gemma3.py +66 -34
- {optimum_rbln-0.8.2a7.dist-info → optimum_rbln-0.8.2rc0.dist-info}/METADATA +1 -1
- {optimum_rbln-0.8.2a7.dist-info → optimum_rbln-0.8.2rc0.dist-info}/RECORD +12 -12
- {optimum_rbln-0.8.2a7.dist-info → optimum_rbln-0.8.2rc0.dist-info}/WHEEL +0 -0
- {optimum_rbln-0.8.2a7.dist-info → optimum_rbln-0.8.2rc0.dist-info}/licenses/LICENSE +0 -0
optimum/rbln/__init__.py
CHANGED
|
@@ -72,6 +72,8 @@ _import_structure = {
|
|
|
72
72
|
"RBLNCLIPVisionModelWithProjectionConfig",
|
|
73
73
|
"RBLNColPaliForRetrieval",
|
|
74
74
|
"RBLNColPaliForRetrievalConfig",
|
|
75
|
+
"RBLNDecoderOnlyModelConfig",
|
|
76
|
+
"RBLNDecoderOnlyModel",
|
|
75
77
|
"RBLNDecoderOnlyModelForCausalLM",
|
|
76
78
|
"RBLNDecoderOnlyModelForCausalLMConfig",
|
|
77
79
|
"RBLNDistilBertForQuestionAnswering",
|
|
@@ -345,6 +347,8 @@ if TYPE_CHECKING:
|
|
|
345
347
|
RBLNCLIPVisionModelWithProjectionConfig,
|
|
346
348
|
RBLNColPaliForRetrieval,
|
|
347
349
|
RBLNColPaliForRetrievalConfig,
|
|
350
|
+
RBLNDecoderOnlyModel,
|
|
351
|
+
RBLNDecoderOnlyModelConfig,
|
|
348
352
|
RBLNDecoderOnlyModelForCausalLM,
|
|
349
353
|
RBLNDecoderOnlyModelForCausalLMConfig,
|
|
350
354
|
RBLNDistilBertForQuestionAnswering,
|
optimum/rbln/__version__.py
CHANGED
|
@@ -17,5 +17,5 @@ __version__: str
|
|
|
17
17
|
__version_tuple__: VERSION_TUPLE
|
|
18
18
|
version_tuple: VERSION_TUPLE
|
|
19
19
|
|
|
20
|
-
__version__ = version = '0.8.
|
|
21
|
-
__version_tuple__ = version_tuple = (0, 8, 2, '
|
|
20
|
+
__version__ = version = '0.8.2rc0'
|
|
21
|
+
__version_tuple__ = version_tuple = (0, 8, 2, 'rc0')
|
|
@@ -62,6 +62,8 @@ _import_structure = {
|
|
|
62
62
|
"RBLNCLIPVisionModelWithProjectionConfig",
|
|
63
63
|
"RBLNDecoderOnlyModelForCausalLM",
|
|
64
64
|
"RBLNDecoderOnlyModelForCausalLMConfig",
|
|
65
|
+
"RBLNDecoderOnlyModelConfig",
|
|
66
|
+
"RBLNDecoderOnlyModel",
|
|
65
67
|
"RBLNDistilBertForQuestionAnswering",
|
|
66
68
|
"RBLNDistilBertForQuestionAnsweringConfig",
|
|
67
69
|
"RBLNDPTForDepthEstimation",
|
|
@@ -196,6 +198,8 @@ if TYPE_CHECKING:
|
|
|
196
198
|
RBLNCLIPVisionModelWithProjectionConfig,
|
|
197
199
|
RBLNColPaliForRetrieval,
|
|
198
200
|
RBLNColPaliForRetrievalConfig,
|
|
201
|
+
RBLNDecoderOnlyModel,
|
|
202
|
+
RBLNDecoderOnlyModelConfig,
|
|
199
203
|
RBLNDecoderOnlyModelForCausalLM,
|
|
200
204
|
RBLNDecoderOnlyModelForCausalLMConfig,
|
|
201
205
|
RBLNDistilBertForQuestionAnswering,
|
|
@@ -84,6 +84,8 @@ _import_structure = {
|
|
|
84
84
|
"RBLNQwen2_5_VLForConditionalGenerationConfig",
|
|
85
85
|
],
|
|
86
86
|
"decoderonly": [
|
|
87
|
+
"RBLNDecoderOnlyModelConfig",
|
|
88
|
+
"RBLNDecoderOnlyModel",
|
|
87
89
|
"RBLNDecoderOnlyModelForCausalLM",
|
|
88
90
|
"RBLNDecoderOnlyModelForCausalLMConfig",
|
|
89
91
|
],
|
|
@@ -216,6 +218,8 @@ if TYPE_CHECKING:
|
|
|
216
218
|
RBLNColPaliForRetrievalConfig,
|
|
217
219
|
)
|
|
218
220
|
from .decoderonly import (
|
|
221
|
+
RBLNDecoderOnlyModel,
|
|
222
|
+
RBLNDecoderOnlyModelConfig,
|
|
219
223
|
RBLNDecoderOnlyModelForCausalLM,
|
|
220
224
|
RBLNDecoderOnlyModelForCausalLMConfig,
|
|
221
225
|
)
|
|
@@ -1025,8 +1025,7 @@ class SlidingWindowAttentionOp(AttentionOp):
|
|
|
1025
1025
|
}
|
|
1026
1026
|
|
|
1027
1027
|
if self.phase == "prefill" or self.phase == "image_prefill":
|
|
1028
|
-
|
|
1029
|
-
op_args["is_bidirectional"] = self.phase == "image_prefill" # FIXME, Hard-coded for Gemma3.
|
|
1028
|
+
op_args["is_bidirectional"] = self.phase == "image_prefill" # FIXME, Hard-coded for Gemma3.
|
|
1030
1029
|
|
|
1031
1030
|
attn_op_name = self.get_attn_op_name()
|
|
1032
1031
|
attn_op = getattr(torch.ops.rbln_custom_ops, attn_op_name, None)
|
|
@@ -1403,7 +1403,7 @@ class RBLNDecoderOnlyModelForCausalLM(RBLNDecoderOnlyModel):
|
|
|
1403
1403
|
)
|
|
1404
1404
|
padded_cache_lengths = torch.zeros_like(generate_idx)
|
|
1405
1405
|
|
|
1406
|
-
#
|
|
1406
|
+
# Prefill
|
|
1407
1407
|
if cache_position is None:
|
|
1408
1408
|
logits = []
|
|
1409
1409
|
inputs = inputs_embeds if inputs_embeds is not None else input_ids
|
|
@@ -23,14 +23,17 @@ class RBLNGemma3ForCausalLMConfig(RBLNDecoderOnlyModelForCausalLMConfig):
|
|
|
23
23
|
self,
|
|
24
24
|
use_position_ids: Optional[bool] = None,
|
|
25
25
|
use_attention_mask: Optional[bool] = None,
|
|
26
|
+
prefill_chunk_size: Optional[int] = None,
|
|
26
27
|
image_prefill_chunk_size: Optional[int] = None,
|
|
27
28
|
**kwargs: Dict[str, Any],
|
|
28
29
|
):
|
|
29
30
|
# use_attention_mask and use_position_ids are always True for Gemma3
|
|
30
31
|
use_attention_mask = use_attention_mask or True
|
|
31
32
|
use_position_ids = use_position_ids or True
|
|
33
|
+
prefill_chunk_size = prefill_chunk_size or 256
|
|
32
34
|
|
|
33
35
|
super().__init__(
|
|
36
|
+
prefill_chunk_size=prefill_chunk_size,
|
|
34
37
|
use_attention_mask=use_attention_mask,
|
|
35
38
|
use_position_ids=use_position_ids,
|
|
36
39
|
**kwargs,
|
|
@@ -73,3 +76,11 @@ class RBLNGemma3ForConditionalGenerationConfig(RBLNModelConfig):
|
|
|
73
76
|
|
|
74
77
|
self.vision_tower = self.init_submodule_config(RBLNSiglipVisionModelConfig, vision_tower)
|
|
75
78
|
self.language_model = self.init_submodule_config(RBLNGemma3ForCausalLMConfig, language_model)
|
|
79
|
+
|
|
80
|
+
@property
|
|
81
|
+
def image_prefill_chunk_size(self):
|
|
82
|
+
return self.language_model.image_prefill_chunk_size
|
|
83
|
+
|
|
84
|
+
@property
|
|
85
|
+
def prefill_chunk_size(self):
|
|
86
|
+
return self.language_model.prefill_chunk_size
|
|
@@ -258,17 +258,45 @@ class RBLNGemma3ForConditionalGeneration(RBLNModel):
|
|
|
258
258
|
|
|
259
259
|
return inputs_embeds
|
|
260
260
|
|
|
261
|
+
def get_padded_cache_position(
|
|
262
|
+
self,
|
|
263
|
+
cache_position: torch.Tensor, # shape: [1, seq_len]
|
|
264
|
+
token_type_ids: torch.Tensor, # shape: [1, seq_len]
|
|
265
|
+
) -> torch.Tensor:
|
|
266
|
+
seq_len = cache_position[0][-1].item() + 1
|
|
267
|
+
|
|
268
|
+
# Find image start positions
|
|
269
|
+
image_starts = [
|
|
270
|
+
s
|
|
271
|
+
for s in torch.where(token_type_ids == 1)[1]
|
|
272
|
+
if torch.all(token_type_ids[:, s : s + self.rbln_config.image_prefill_chunk_size] == 1)
|
|
273
|
+
]
|
|
274
|
+
|
|
275
|
+
# Initialize padded tensors
|
|
276
|
+
padded_input_len = seq_len
|
|
277
|
+
for image_start in image_starts:
|
|
278
|
+
pad_needed = (
|
|
279
|
+
self.rbln_config.image_prefill_chunk_size
|
|
280
|
+
- (image_start + padded_input_len - seq_len) % self.rbln_config.image_prefill_chunk_size
|
|
281
|
+
) % self.rbln_config.image_prefill_chunk_size
|
|
282
|
+
padded_input_len += pad_needed
|
|
283
|
+
|
|
284
|
+
return torch.cat(
|
|
285
|
+
[cache_position, torch.arange(seq_len, padded_input_len, dtype=torch.int32).unsqueeze(0)],
|
|
286
|
+
dim=1,
|
|
287
|
+
)
|
|
288
|
+
|
|
261
289
|
def forward(
|
|
262
290
|
self,
|
|
263
291
|
input_ids: torch.LongTensor = None,
|
|
292
|
+
attention_mask: torch.Tensor = None,
|
|
293
|
+
token_type_ids: torch.Tensor = None,
|
|
264
294
|
pixel_values: torch.FloatTensor = None,
|
|
265
|
-
attention_mask: Optional[torch.Tensor] = None,
|
|
266
295
|
cache_position: Optional[torch.LongTensor] = None,
|
|
267
296
|
inputs_embeds: Optional[torch.FloatTensor] = None,
|
|
268
297
|
generate_idx: Optional[torch.Tensor] = None,
|
|
269
298
|
padded_cache_lengths: Optional[torch.Tensor] = None,
|
|
270
299
|
position_ids: Optional[torch.Tensor] = None,
|
|
271
|
-
token_type_ids: Optional[torch.Tensor] = None,
|
|
272
300
|
**lm_kwargs: Dict[str, Any],
|
|
273
301
|
) -> Union[Tuple, RBLNDecoderOnlyForCausalLMOutput]:
|
|
274
302
|
# prefill
|
|
@@ -279,12 +307,15 @@ class RBLNGemma3ForConditionalGeneration(RBLNModel):
|
|
|
279
307
|
|
|
280
308
|
for b_idx in range(batch_size):
|
|
281
309
|
cache_position = torch.arange(0, generate_idx[b_idx].item(), dtype=torch.int32).unsqueeze(0)
|
|
310
|
+
token_type_id = token_type_ids[b_idx : b_idx + 1, attention_mask[b_idx].bool()]
|
|
311
|
+
cache_position = self.get_padded_cache_position(cache_position, token_type_id)
|
|
312
|
+
|
|
282
313
|
output = self.language_model.prefill_decoder(
|
|
283
314
|
inputs_embeds=inputs_embeds[b_idx : b_idx + 1],
|
|
284
315
|
attention_mask=attention_mask[b_idx],
|
|
285
316
|
cache_position=cache_position,
|
|
286
317
|
batch_idx=b_idx,
|
|
287
|
-
token_type_ids=token_type_ids[b_idx : b_idx + 1]
|
|
318
|
+
token_type_ids=token_type_ids[b_idx : b_idx + 1], # do not pass token_type_id
|
|
288
319
|
)
|
|
289
320
|
padded_cache_lengths[b_idx] += output.padded_cache_lengths
|
|
290
321
|
logits.append(output.logits)
|
|
@@ -336,9 +367,10 @@ class RBLNGemma3RuntimeModel(RBLNRuntimeModel):
|
|
|
336
367
|
# chunked_attention_mask shape
|
|
337
368
|
chunked_attention_mask = torch.zeros(1, chunked_attention_mask.shape[-1], dtype=torch.float32)
|
|
338
369
|
|
|
339
|
-
#
|
|
370
|
+
# In case of Gemma3ForConditionalGeneration, the loop counter may not be a prefill_chunk_size,
|
|
371
|
+
# so we cannot guarantee that the last chunk starts at a position that is a multiple of prefill_chunk_size.
|
|
340
372
|
if self.rbln_config.use_image_prefill:
|
|
341
|
-
padding_size =
|
|
373
|
+
padding_size = self.rbln_config.image_prefill_chunk_size
|
|
342
374
|
inputs = torch.nn.functional.pad(inputs, (0, 0, 0, padding_size))
|
|
343
375
|
cache_position = torch.nn.functional.pad(cache_position, (0, padding_size))
|
|
344
376
|
position_ids = torch.nn.functional.pad(position_ids, (0, padding_size))
|
|
@@ -389,45 +421,39 @@ class RBLNGemma3RuntimeModel(RBLNRuntimeModel):
|
|
|
389
421
|
|
|
390
422
|
step = 0
|
|
391
423
|
while step < query_length:
|
|
392
|
-
|
|
393
|
-
|
|
394
|
-
|
|
395
|
-
|
|
396
|
-
|
|
397
|
-
|
|
398
|
-
|
|
399
|
-
|
|
400
|
-
|
|
401
|
-
|
|
402
|
-
|
|
403
|
-
and not is_image_prefill
|
|
404
|
-
and torch.any(token_type_ids[:, step : step + prefill_chunk_size] == 1)
|
|
405
|
-
)
|
|
406
|
-
|
|
407
|
-
# Check if the prefill chunk crosses a block boundary, requiring padding to align with block boundaries
|
|
408
|
-
is_cross_block_boundary = (
|
|
409
|
-
step // self.rbln_config.kvcache_block_size
|
|
410
|
-
!= (step + prefill_chunk_size) // self.rbln_config.kvcache_block_size
|
|
411
|
-
)
|
|
424
|
+
if self.rbln_config.use_image_prefill:
|
|
425
|
+
# Check if the prefill chunk is an image prefill
|
|
426
|
+
is_image_prefill = torch.all(
|
|
427
|
+
token_type_ids[:, step : step + self.rbln_config.image_prefill_chunk_size] == 1
|
|
428
|
+
)
|
|
429
|
+
# Check if the prefill chunk is a text prefill which have image_tokens in it.
|
|
430
|
+
is_text_prefill_with_image_tokens = not is_image_prefill and torch.any(
|
|
431
|
+
token_type_ids[:, step : step + self.rbln_config.prefill_chunk_size] == 1
|
|
432
|
+
)
|
|
433
|
+
else:
|
|
434
|
+
is_image_prefill, is_text_prefill_with_image_tokens = False, False
|
|
412
435
|
|
|
413
436
|
# Check if the prefill chunk is the last chunk
|
|
414
|
-
is_last_chunk = step + prefill_chunk_size >= query_length
|
|
437
|
+
is_last_chunk = step + self.rbln_config.prefill_chunk_size >= query_length
|
|
415
438
|
|
|
416
|
-
|
|
417
|
-
|
|
418
|
-
|
|
439
|
+
input_chunk = inputs[:, step : step + self.rbln_config.prefill_chunk_size]
|
|
440
|
+
cache_pos_chunk = (
|
|
441
|
+
cache_position[:, step : step + self.rbln_config.prefill_chunk_size] + padded_cache_lengths
|
|
442
|
+
)
|
|
443
|
+
position_ids_chunk = position_ids[:, step : step + self.rbln_config.prefill_chunk_size]
|
|
419
444
|
|
|
420
445
|
# if text_prefill end with image_tokens, we only treat the text part.
|
|
421
|
-
num_processed_tokens = prefill_chunk_size
|
|
446
|
+
num_processed_tokens = self.rbln_config.prefill_chunk_size
|
|
447
|
+
current_padded_cache_lengths = 0
|
|
422
448
|
if is_text_prefill_with_image_tokens:
|
|
423
|
-
first_image_token_idx = torch.where(
|
|
449
|
+
first_image_token_idx = torch.where(
|
|
450
|
+
token_type_ids[:, step : step + self.rbln_config.prefill_chunk_size] == 1
|
|
451
|
+
)[1][0]
|
|
424
452
|
num_processed_tokens = first_image_token_idx.item()
|
|
453
|
+
current_padded_cache_lengths = self.rbln_config.prefill_chunk_size - num_processed_tokens
|
|
425
454
|
if is_last_chunk:
|
|
426
455
|
num_processed_tokens = query_length - step
|
|
427
456
|
|
|
428
|
-
input_chunk = inputs[:, step : step + prefill_chunk_size]
|
|
429
|
-
cache_pos_chunk = cache_position[:, step : step + prefill_chunk_size].clone() + padded_cache_lengths
|
|
430
|
-
position_ids_chunk = position_ids[:, step : step + prefill_chunk_size].clone()
|
|
431
457
|
chunked_attention_mask[
|
|
432
458
|
:, step + padded_cache_lengths : step + num_processed_tokens + padded_cache_lengths
|
|
433
459
|
] = 1
|
|
@@ -456,6 +482,7 @@ class RBLNGemma3RuntimeModel(RBLNRuntimeModel):
|
|
|
456
482
|
out=out_buffers,
|
|
457
483
|
)
|
|
458
484
|
|
|
485
|
+
padded_cache_lengths += current_padded_cache_lengths
|
|
459
486
|
step += num_processed_tokens
|
|
460
487
|
|
|
461
488
|
if not is_external_block_tables:
|
|
@@ -633,6 +660,11 @@ class RBLNGemma3ForCausalLM(RBLNDecoderOnlyModelForCausalLM):
|
|
|
633
660
|
raise ValueError("use_attention_mask and use_position_ids must be True for RBLNGemma3ForCausalLM")
|
|
634
661
|
|
|
635
662
|
if rbln_config.use_image_prefill:
|
|
663
|
+
if rbln_config.prefill_chunk_size != rbln_config.image_prefill_chunk_size:
|
|
664
|
+
raise NotImplementedError(
|
|
665
|
+
"Not implemented for different prefill chunk sizes between text and image prefill."
|
|
666
|
+
)
|
|
667
|
+
|
|
636
668
|
# Update image prefill compile config
|
|
637
669
|
img_prefill_input_info = cls.get_input_info(
|
|
638
670
|
batch_size=1,
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: optimum-rbln
|
|
3
|
-
Version: 0.8.
|
|
3
|
+
Version: 0.8.2rc0
|
|
4
4
|
Summary: Optimum RBLN is the interface between the HuggingFace Transformers and Diffusers libraries and RBLN accelerators. It provides a set of tools enabling easy model loading and inference on single and multiple rbln device settings for different downstream tasks.
|
|
5
5
|
Project-URL: Homepage, https://rebellions.ai
|
|
6
6
|
Project-URL: Documentation, https://docs.rbln.ai
|
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
optimum/rbln/__init__.py,sha256=
|
|
2
|
-
optimum/rbln/__version__.py,sha256=
|
|
1
|
+
optimum/rbln/__init__.py,sha256=4iR9FOTeDLiRU2x_c30KqNOSMvQwA3OqZVbZDOefDHE,17120
|
|
2
|
+
optimum/rbln/__version__.py,sha256=dgBEG8WaV-ceU4n4KOOPGcemoEt9d8J8ArR9VIbipXo,521
|
|
3
3
|
optimum/rbln/configuration_utils.py,sha256=FluYXTBB3C93-_35Z_XSdVnN6ScIj4AuKt-4QFt0m8g,33116
|
|
4
4
|
optimum/rbln/modeling.py,sha256=gww-H-q16_mGw2qGnFwOjEj3J9yMjBKnRTKlnCkVlx8,14315
|
|
5
5
|
optimum/rbln/modeling_base.py,sha256=AShxAt3KIOcCqfyF4U83dIrKwoj4p2Kxtc1ns_9-ltU,24154
|
|
@@ -71,12 +71,12 @@ optimum/rbln/ops/flash_attn.py,sha256=z39DJZSk94630ueoOCkiybxR5gzvNR-SRADHs0F6pz
|
|
|
71
71
|
optimum/rbln/ops/kv_cache_update.py,sha256=aIvK2Sp7M3EfJzJgNvIvAJv4emoN6QOhmgaWj-VboLs,1440
|
|
72
72
|
optimum/rbln/ops/linear.py,sha256=5K3pcrrUHu_p8LrMIU-jX2TnafksveFjjZSCsYSp_yw,1328
|
|
73
73
|
optimum/rbln/ops/sliding_window_attn.py,sha256=EQrV_yRGc5z6kvwEsAcLP028bJWkQg2UPI3xubt9skU,3487
|
|
74
|
-
optimum/rbln/transformers/__init__.py,sha256=
|
|
74
|
+
optimum/rbln/transformers/__init__.py,sha256=ioH-OpCtBsyOVfXz-6ZY0tuGzHjwWn1kfgOP60pcTIc,11049
|
|
75
75
|
optimum/rbln/transformers/configuration_generic.py,sha256=kNhPWtzF0IovUnrsXfxXdXITqgpfCAAedjfB6jSAhEg,5131
|
|
76
76
|
optimum/rbln/transformers/modeling_attention_utils.py,sha256=aLyOaq4me1m-JMmnKbuyNQageDxNU2jjEhGE_ew2P5o,11465
|
|
77
77
|
optimum/rbln/transformers/modeling_generic.py,sha256=SXsZghRDsPolNnG5FFPRtXzIEmPQDnz0iRy6PIZvFVI,12225
|
|
78
78
|
optimum/rbln/transformers/modeling_rope_utils.py,sha256=6Zg3r-TeUk4WQAlr95pqfhuoAD_RQ4njT1rbO9uPL0Q,14379
|
|
79
|
-
optimum/rbln/transformers/models/__init__.py,sha256=
|
|
79
|
+
optimum/rbln/transformers/models/__init__.py,sha256=rv-4CyM-NuC-m_B-G4kYu3fh_xDmOgkHW5B_cIu-M8o,11929
|
|
80
80
|
optimum/rbln/transformers/models/audio_spectrogram_transformer/__init__.py,sha256=I2vL4lrzbT5p4eJcH-EKHzEfcPkj_XVsie7jb9q6yic,775
|
|
81
81
|
optimum/rbln/transformers/models/audio_spectrogram_transformer/configuration_audio_spectrogram_transformer.py,sha256=z7LJiVJPmnlCM3mcyhPJP8AufSrxO_dsPeJ51onq-Nc,833
|
|
82
82
|
optimum/rbln/transformers/models/audio_spectrogram_transformer/modeling_audio_spectrogram_transformer.py,sha256=FIKEVWpIt6-JQX9B_rAfCrAPqdUHtR2i8D_X2k7639E,1498
|
|
@@ -102,8 +102,8 @@ optimum/rbln/transformers/models/colpali/configuration_colpali.py,sha256=ieY-tuy
|
|
|
102
102
|
optimum/rbln/transformers/models/colpali/modeling_colpali.py,sha256=jzvJCBrrCXSpjfmJ3O-VvPNFGWGaNbpOV09JwLPAZWs,15757
|
|
103
103
|
optimum/rbln/transformers/models/decoderonly/__init__.py,sha256=w3VZOIBYaHXVdnuhK4y0zWAj0IAv7_5LGTJYaz9oYmI,1056
|
|
104
104
|
optimum/rbln/transformers/models/decoderonly/configuration_decoderonly.py,sha256=tZNT0kz90V4xVDJ9I_9WjPZjLHwhFZpKshija64vcMA,23868
|
|
105
|
-
optimum/rbln/transformers/models/decoderonly/decoderonly_architecture.py,sha256=
|
|
106
|
-
optimum/rbln/transformers/models/decoderonly/modeling_decoderonly.py,sha256=
|
|
105
|
+
optimum/rbln/transformers/models/decoderonly/decoderonly_architecture.py,sha256=jSVljf4dUpAzbTQSfpl1a0WTQ2wyR2k-X51bHw20Beg,42133
|
|
106
|
+
optimum/rbln/transformers/models/decoderonly/modeling_decoderonly.py,sha256=8YaRfQTEAlLspVQzxJ3lp-2UKI3pye3WncZYC-VbQzo,62427
|
|
107
107
|
optimum/rbln/transformers/models/distilbert/__init__.py,sha256=zXL78SOEORTnUN_wrdoaDaYpntG8lcFHvPobM6jC0CI,841
|
|
108
108
|
optimum/rbln/transformers/models/distilbert/configuration_distilbert.py,sha256=O3BW9JjyYk9PLyiofvOKEgTdMZ_jpIuPfot281pSsyg,984
|
|
109
109
|
optimum/rbln/transformers/models/distilbert/modeling_distilbert.py,sha256=LUh6zYGa8AR3Yxaj3gtyJRc-czBN3qnHTc-JTAhuqY0,1099
|
|
@@ -119,9 +119,9 @@ optimum/rbln/transformers/models/gemma/configuration_gemma.py,sha256=H1nVp8HBJxx
|
|
|
119
119
|
optimum/rbln/transformers/models/gemma/gemma_architecture.py,sha256=2Ivay8NTSHmQAqXFh9JvG6Ja5rMThcRAjYPzyipcRI8,956
|
|
120
120
|
optimum/rbln/transformers/models/gemma/modeling_gemma.py,sha256=p9mxRkxu8xAa3g-v62-A0mbcfiQrLIepw2R3BpZP3Cg,3659
|
|
121
121
|
optimum/rbln/transformers/models/gemma3/__init__.py,sha256=6rugk3615SEt4lh7gduo_J9VyGiSReuEIvL0Uno0eaI,790
|
|
122
|
-
optimum/rbln/transformers/models/gemma3/configuration_gemma3.py,sha256=
|
|
122
|
+
optimum/rbln/transformers/models/gemma3/configuration_gemma3.py,sha256=GdKyPbsA6zBdr5YsKodLpPcOwG2fCwb_VaXfgnf7j7E,3426
|
|
123
123
|
optimum/rbln/transformers/models/gemma3/gemma3_architecture.py,sha256=fpLDAXCe5paWVsfc0tL59JkRQMRF-WNgIzOIb_QpSLU,6191
|
|
124
|
-
optimum/rbln/transformers/models/gemma3/modeling_gemma3.py,sha256=
|
|
124
|
+
optimum/rbln/transformers/models/gemma3/modeling_gemma3.py,sha256=0eJAwOBOdE5LttS3kECwcldzVDHbrRVhw2Ycx29274s,34079
|
|
125
125
|
optimum/rbln/transformers/models/gpt2/__init__.py,sha256=SsawHMStE3wYRtqkH5EvdTFkCdX0LLmp-QSKFhEBrHo,740
|
|
126
126
|
optimum/rbln/transformers/models/gpt2/configuration_gpt2.py,sha256=iGdHfzG7plekZcIz-Z5U8lRE4SB8gbJJNcFQJ9l8Myg,1533
|
|
127
127
|
optimum/rbln/transformers/models/gpt2/gpt2_architecture.py,sha256=MyAWReXmyuHnDpW5HI_TI7psyJZxLujZ9KT5XnNm7nA,2802
|
|
@@ -222,7 +222,7 @@ optimum/rbln/utils/model_utils.py,sha256=4k5879Kh75m3x_vS4-qOGfqsOiAvc2kdNFFfvsF
|
|
|
222
222
|
optimum/rbln/utils/runtime_utils.py,sha256=nIJioiN16nAyAzoArKjsy5ocLUsrr0UEy4f3LNT82SA,7961
|
|
223
223
|
optimum/rbln/utils/save_utils.py,sha256=hG5uOtYmecSXZuGTvCXsTM-SiyZpr5q3InUGCCq_jzQ,3619
|
|
224
224
|
optimum/rbln/utils/submodule.py,sha256=w5mgPgncI740gVKMu3S-69DGNdUSI0bTZxegQGcZ98Y,5011
|
|
225
|
-
optimum_rbln-0.8.
|
|
226
|
-
optimum_rbln-0.8.
|
|
227
|
-
optimum_rbln-0.8.
|
|
228
|
-
optimum_rbln-0.8.
|
|
225
|
+
optimum_rbln-0.8.2rc0.dist-info/METADATA,sha256=INTkgEccxOUtO4bwt14hbQz8eWLG6REbSC7LJtVmAoQ,5300
|
|
226
|
+
optimum_rbln-0.8.2rc0.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
|
|
227
|
+
optimum_rbln-0.8.2rc0.dist-info/licenses/LICENSE,sha256=QwcOLU5TJoTeUhuIXzhdCEEDDvorGiC6-3YTOl4TecE,11356
|
|
228
|
+
optimum_rbln-0.8.2rc0.dist-info/RECORD,,
|
|
File without changes
|
|
File without changes
|