nexaai 1.0.19rc7__cp310-cp310-macosx_13_0_x86_64.whl → 1.0.19rc9__cp310-cp310-macosx_13_0_x86_64.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 nexaai might be problematic. Click here for more details.
- nexaai/_stub.cpython-310-darwin.so +0 -0
- nexaai/_version.py +1 -1
- nexaai/binds/libnexa_bridge.dylib +0 -0
- nexaai/mlx_backend/vlm/generate_qwen3_vl.py +14 -31
- nexaai/mlx_backend/vlm/generate_qwen3_vl_moe.py +15 -32
- nexaai/mlx_backend/vlm/modeling/models/qwen3_vl/qwen3vl.py +7 -23
- nexaai/mlx_backend/vlm/modeling/models/qwen3vl_moe/qwen3vl_moe.py +8 -24
- {nexaai-1.0.19rc7.dist-info → nexaai-1.0.19rc9.dist-info}/METADATA +1 -1
- {nexaai-1.0.19rc7.dist-info → nexaai-1.0.19rc9.dist-info}/RECORD +11 -11
- {nexaai-1.0.19rc7.dist-info → nexaai-1.0.19rc9.dist-info}/WHEEL +0 -0
- {nexaai-1.0.19rc7.dist-info → nexaai-1.0.19rc9.dist-info}/top_level.txt +0 -0
|
Binary file
|
nexaai/_version.py
CHANGED
|
Binary file
|
|
@@ -1,6 +1,5 @@
|
|
|
1
1
|
import argparse
|
|
2
2
|
import json
|
|
3
|
-
import sys
|
|
4
3
|
import os
|
|
5
4
|
import mlx.core as mx
|
|
6
5
|
import mlx.nn as nn
|
|
@@ -10,38 +9,21 @@ import requests
|
|
|
10
9
|
import numpy as np
|
|
11
10
|
from pathlib import Path
|
|
12
11
|
from huggingface_hub import snapshot_download
|
|
13
|
-
|
|
14
|
-
|
|
15
|
-
curr_dir = os.path.dirname(os.path.abspath(__file__))
|
|
16
|
-
sys.path.append(curr_dir)
|
|
17
|
-
sys.path.append(os.path.dirname(curr_dir))
|
|
18
|
-
|
|
19
|
-
# Add the qwen3vl model directory to path
|
|
20
|
-
qwen3vl_dir = os.path.join(curr_dir, "modeling", "models", "qwen3_vl")
|
|
21
|
-
sys.path.append(qwen3vl_dir)
|
|
12
|
+
from dataclasses import dataclass
|
|
13
|
+
from typing import Any, Generator, List, Optional, Sequence, Tuple, Union
|
|
22
14
|
|
|
23
15
|
# Import required modules for quantized loading
|
|
24
16
|
from transformers import AutoTokenizer
|
|
25
17
|
|
|
26
|
-
#
|
|
27
|
-
|
|
28
|
-
|
|
29
|
-
|
|
30
|
-
|
|
31
|
-
|
|
32
|
-
|
|
33
|
-
from .modeling.models.qwen3_vl.processor import Qwen3VLProcessor
|
|
34
|
-
except ImportError:
|
|
35
|
-
# Fallback for Nuitka compiled environment - use sys.path approach
|
|
36
|
-
from llm_common.generate import nexa_generate_step
|
|
37
|
-
from llm_common.cache import make_prompt_cache
|
|
38
|
-
from qwen3vl import VEGModel, LLMModel, ModelArgs, VisionConfig, TextConfig, handle_multimodal_embeds
|
|
39
|
-
from processor import Qwen3VLProcessor
|
|
40
|
-
|
|
41
|
-
from ml import ChatMessage
|
|
42
|
-
from dataclasses import dataclass
|
|
43
|
-
from typing import Any, Generator, List, Optional, Sequence, Tuple, Union
|
|
18
|
+
# Import from the nested modeling structure
|
|
19
|
+
from .modeling.models.qwen3_vl.llm_common.generate import nexa_generate_step
|
|
20
|
+
from .modeling.models.qwen3_vl.llm_common.cache import make_prompt_cache
|
|
21
|
+
from .modeling.models.qwen3_vl.qwen3vl import (
|
|
22
|
+
VEGModel, LLMModel, ModelArgs, VisionConfig, TextConfig, handle_multimodal_embeds
|
|
23
|
+
)
|
|
24
|
+
from .modeling.models.qwen3_vl.processor import Qwen3VLProcessor
|
|
44
25
|
from .generate import GenerationResult
|
|
26
|
+
from ml import ChatMessage
|
|
45
27
|
|
|
46
28
|
# Custom exception for context length exceeded
|
|
47
29
|
class ContextLengthExceededError(Exception):
|
|
@@ -79,10 +61,11 @@ def load_qwen3_vl(
|
|
|
79
61
|
model_path = Path(snapshot_download(
|
|
80
62
|
repo_id=path_or_repo, repo_type="model", revision=revision))
|
|
81
63
|
else:
|
|
82
|
-
# Fallback to local modelfiles directory
|
|
83
|
-
|
|
64
|
+
# Fallback to local modelfiles directory relative to this file
|
|
65
|
+
curr_dir = Path(__file__).parent
|
|
66
|
+
model_path = curr_dir / "modeling" / "models" / "qwen3_vl" / "modelfiles"
|
|
84
67
|
if not model_path.exists():
|
|
85
|
-
model_path =
|
|
68
|
+
model_path = curr_dir / "modelfiles"
|
|
86
69
|
|
|
87
70
|
# Model configs (kept identical to main)
|
|
88
71
|
vision_config = VisionConfig(
|
|
@@ -1,6 +1,5 @@
|
|
|
1
1
|
import argparse
|
|
2
2
|
import json
|
|
3
|
-
import sys
|
|
4
3
|
import os
|
|
5
4
|
import mlx.core as mx
|
|
6
5
|
import mlx.nn as nn
|
|
@@ -10,38 +9,21 @@ import requests
|
|
|
10
9
|
import numpy as np
|
|
11
10
|
from pathlib import Path
|
|
12
11
|
from huggingface_hub import snapshot_download
|
|
13
|
-
|
|
14
|
-
|
|
15
|
-
curr_dir = os.path.dirname(os.path.abspath(__file__))
|
|
16
|
-
sys.path.append(curr_dir)
|
|
17
|
-
sys.path.append(os.path.dirname(curr_dir))
|
|
18
|
-
|
|
19
|
-
# Add the qwen3vl model directory to path
|
|
20
|
-
qwen3vl_dir = os.path.join(curr_dir, "modeling", "models", "qwen3vl_moe")
|
|
21
|
-
sys.path.append(qwen3vl_dir)
|
|
12
|
+
from dataclasses import dataclass
|
|
13
|
+
from typing import Any, Generator, List, Optional, Sequence, Tuple, Union
|
|
22
14
|
|
|
23
15
|
# Import required modules for quantized loading
|
|
24
16
|
from transformers import AutoTokenizer
|
|
25
17
|
|
|
26
|
-
#
|
|
27
|
-
|
|
28
|
-
|
|
29
|
-
|
|
30
|
-
|
|
31
|
-
|
|
32
|
-
|
|
33
|
-
|
|
34
|
-
|
|
35
|
-
# Fallback for Nuitka compiled environment - use sys.path approach
|
|
36
|
-
from llm_common.generate import nexa_generate_step
|
|
37
|
-
from llm_common.cache import make_prompt_cache
|
|
38
|
-
from qwen3vl_moe import VEGModel, LLMModel, ModelArgs, VisionConfig, TextConfig, handle_multimodal_embeds
|
|
39
|
-
from processor import Qwen3VLProcessor
|
|
40
|
-
|
|
41
|
-
from ml import ChatMessage
|
|
42
|
-
from dataclasses import dataclass
|
|
43
|
-
from typing import Any, Generator, List, Optional, Sequence, Tuple, Union
|
|
44
|
-
from .generate import GenerationResult
|
|
18
|
+
# Import from the nested modeling structure
|
|
19
|
+
from .modeling.models.qwen3vl_moe.llm_common.generate import nexa_generate_step
|
|
20
|
+
from .modeling.models.qwen3vl_moe.llm_common.cache import make_prompt_cache
|
|
21
|
+
from .modeling.models.qwen3vl_moe.qwen3vl_moe import (
|
|
22
|
+
VEGModel, LLMModel, ModelArgs, VisionConfig, TextConfig, handle_multimodal_embeds
|
|
23
|
+
)
|
|
24
|
+
from .modeling.models.qwen3vl_moe.processor import Qwen3VLProcessor
|
|
25
|
+
from .generate import GenerationResult
|
|
26
|
+
from ml import ChatMessage
|
|
45
27
|
|
|
46
28
|
@dataclass
|
|
47
29
|
class Qwen3VLBundledModel:
|
|
@@ -73,10 +55,11 @@ def load_qwen3_vl(
|
|
|
73
55
|
model_path = Path(snapshot_download(
|
|
74
56
|
repo_id=path_or_repo, repo_type="model", revision=revision))
|
|
75
57
|
else:
|
|
76
|
-
# Fallback to local modelfiles directory
|
|
77
|
-
|
|
58
|
+
# Fallback to local modelfiles directory relative to this file
|
|
59
|
+
curr_dir = Path(__file__).parent
|
|
60
|
+
model_path = curr_dir / "modeling" / "models" / "qwen3vl_moe" / "modelfiles"
|
|
78
61
|
if not model_path.exists():
|
|
79
|
-
model_path =
|
|
62
|
+
model_path = curr_dir / "modelfiles"
|
|
80
63
|
|
|
81
64
|
# Model configs - Updated to match Qwen3VL-MoE specifications
|
|
82
65
|
vision_config = VisionConfig(
|
|
@@ -8,29 +8,13 @@ import mlx.nn as nn
|
|
|
8
8
|
import math
|
|
9
9
|
import numpy as np
|
|
10
10
|
|
|
11
|
-
|
|
12
|
-
import
|
|
13
|
-
|
|
14
|
-
|
|
15
|
-
|
|
16
|
-
|
|
17
|
-
|
|
18
|
-
# Try relative imports first, fallback to sys.path approach for Nuitka compatibility
|
|
19
|
-
try:
|
|
20
|
-
from .llm_common.base import (
|
|
21
|
-
BaseModelArgs,
|
|
22
|
-
create_attention_mask,
|
|
23
|
-
scaled_dot_product_attention,
|
|
24
|
-
)
|
|
25
|
-
from .llm_common.rope_utils import initialize_rope
|
|
26
|
-
except ImportError:
|
|
27
|
-
# Fallback for Nuitka compiled environment
|
|
28
|
-
from llm_common.base import (
|
|
29
|
-
BaseModelArgs,
|
|
30
|
-
create_attention_mask,
|
|
31
|
-
scaled_dot_product_attention,
|
|
32
|
-
)
|
|
33
|
-
from llm_common.rope_utils import initialize_rope
|
|
11
|
+
# Import from nested llm_common structure using relative imports
|
|
12
|
+
from .llm_common.base import (
|
|
13
|
+
BaseModelArgs,
|
|
14
|
+
create_attention_mask,
|
|
15
|
+
scaled_dot_product_attention,
|
|
16
|
+
)
|
|
17
|
+
from .llm_common.rope_utils import initialize_rope
|
|
34
18
|
|
|
35
19
|
|
|
36
20
|
@dataclass
|
|
@@ -8,30 +8,14 @@ import mlx.nn as nn
|
|
|
8
8
|
import math
|
|
9
9
|
import numpy as np
|
|
10
10
|
|
|
11
|
-
|
|
12
|
-
import
|
|
13
|
-
|
|
14
|
-
|
|
15
|
-
|
|
16
|
-
|
|
17
|
-
|
|
18
|
-
|
|
19
|
-
try:
|
|
20
|
-
from .llm_common.base import (
|
|
21
|
-
BaseModelArgs,
|
|
22
|
-
create_attention_mask,
|
|
23
|
-
scaled_dot_product_attention,
|
|
24
|
-
)
|
|
25
|
-
from .llm_common.rope_utils import initialize_rope
|
|
26
|
-
except ImportError:
|
|
27
|
-
# Fallback for Nuitka compiled environment
|
|
28
|
-
from llm_common.base import (
|
|
29
|
-
BaseModelArgs,
|
|
30
|
-
create_attention_mask,
|
|
31
|
-
scaled_dot_product_attention,
|
|
32
|
-
)
|
|
33
|
-
from llm_common.rope_utils import initialize_rope
|
|
34
|
-
from switch_layers import SwitchGLU
|
|
11
|
+
# Import from nested llm_common structure using relative imports
|
|
12
|
+
from .llm_common.base import (
|
|
13
|
+
BaseModelArgs,
|
|
14
|
+
create_attention_mask,
|
|
15
|
+
scaled_dot_product_attention,
|
|
16
|
+
)
|
|
17
|
+
from .llm_common.rope_utils import initialize_rope
|
|
18
|
+
from .switch_layers import SwitchGLU
|
|
35
19
|
|
|
36
20
|
|
|
37
21
|
@dataclass
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
nexaai/__init__.py,sha256=L8oB7GFZZMGnUpCg0PecDbI_ycKuQak-ZEJ4Y12_QIw,2184
|
|
2
|
-
nexaai/_stub.cpython-310-darwin.so,sha256=
|
|
3
|
-
nexaai/_version.py,sha256=
|
|
2
|
+
nexaai/_stub.cpython-310-darwin.so,sha256=X4Ayzpy-yKcxfF2Ag2w6V0jZuOMGavnB16kgBJJKXnU,49832
|
|
3
|
+
nexaai/_version.py,sha256=cBq07C7RkT85kn1w5XjXDE5IIdTu6Lg8iJ-tx2W_xdI,143
|
|
4
4
|
nexaai/asr.py,sha256=NljMXDErwPNMOPaRkJZMEDka9Nk8xyur7L8i924TStY,2054
|
|
5
5
|
nexaai/base.py,sha256=N8PRgDFA-XPku2vWnQIofQ7ipz3pPlO6f8YZGnuhquE,982
|
|
6
6
|
nexaai/common.py,sha256=Y0NJNLTi4Nq4x1WL6PQsSvGUto0eGmWhjpsC6jcekfA,3444
|
|
@@ -19,7 +19,7 @@ nexaai/asr_impl/pybind_asr_impl.py,sha256=pE9Hb_hMi5yAc4MF83bLVOb8zDtreCkB3_u7XE
|
|
|
19
19
|
nexaai/binds/__init__.py,sha256=eYuay_8DDXeOUWz2_R9HFSabohxs6hvZn391t2L0Po0,104
|
|
20
20
|
nexaai/binds/common_bind.cpython-310-darwin.so,sha256=BoXByRlNGDaNS1YyZyCF-s7h0vXP9NLPlJMQQ5pqusU,235488
|
|
21
21
|
nexaai/binds/embedder_bind.cpython-310-darwin.so,sha256=b2NoXFAJvPLi_P1X7lXLKmAUU0v2HJI3Zwa10gfqHdw,202032
|
|
22
|
-
nexaai/binds/libnexa_bridge.dylib,sha256
|
|
22
|
+
nexaai/binds/libnexa_bridge.dylib,sha256=kFyl0jt6NRkoJsRtz5hwZmfBy4eOnOFLoaHmJb3hupQ,251224
|
|
23
23
|
nexaai/binds/llm_bind.cpython-310-darwin.so,sha256=p1ZTGMolEkWywkmwzOUjTr3RpSEH21BHZAggVzo89Ks,183088
|
|
24
24
|
nexaai/binds/vlm_bind.cpython-310-darwin.so,sha256=LGd-tykePnQFfGca25HnPIBfXsfrMzbwyx6d5Ld3xps,183000
|
|
25
25
|
nexaai/binds/nexa_llama_cpp/libggml-base.dylib,sha256=lARxdlzBWnBjYWAuQQF8utDbj_kX-J7RNxjrCiOQd2M,629760
|
|
@@ -246,8 +246,8 @@ nexaai/mlx_backend/tts/__init__.py,sha256=fuT_9_xpYJ28m4yjly5L2jChUrzlSQz-b_S7nu
|
|
|
246
246
|
nexaai/mlx_backend/tts/interface.py,sha256=0FvZbIyOvg8jERZEQ6bygbv7v02O9xHO4-TPUlar0b4,9568
|
|
247
247
|
nexaai/mlx_backend/vlm/__init__.py,sha256=_25kvMEviX16Hg3bro8Ws70V0eeIEqYKV8ZDXqYzKew,73
|
|
248
248
|
nexaai/mlx_backend/vlm/generate.py,sha256=DqHFEAuqk-nko8ho6U9GAXTDAWz4d8GTe_hCt-XFyCw,19071
|
|
249
|
-
nexaai/mlx_backend/vlm/generate_qwen3_vl.py,sha256=
|
|
250
|
-
nexaai/mlx_backend/vlm/generate_qwen3_vl_moe.py,sha256=
|
|
249
|
+
nexaai/mlx_backend/vlm/generate_qwen3_vl.py,sha256=NaVOU1kKRpkCClN5yV3_iL4aIe0V9LwqoKx-HI_xDNg,9590
|
|
250
|
+
nexaai/mlx_backend/vlm/generate_qwen3_vl_moe.py,sha256=ZSbM8JjTlkxUaVO9UNZM6YSbd60am3Z4ztJJEBsnJHg,9015
|
|
251
251
|
nexaai/mlx_backend/vlm/interface.py,sha256=uTX5ifnMltDiFu5T4OgHCPNHf0E4gGUEFR8Bz-jnbds,19774
|
|
252
252
|
nexaai/mlx_backend/vlm/main.py,sha256=pvNrMzgaYxw-uyQLj1k-OiAI6RdmNVxhxemEMXZSwUs,10829
|
|
253
253
|
nexaai/mlx_backend/vlm/modeling/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
|
@@ -359,7 +359,7 @@ nexaai/mlx_backend/vlm/modeling/models/qwen2_vl/language.py,sha256=OeUoAfIg7dNB4
|
|
|
359
359
|
nexaai/mlx_backend/vlm/modeling/models/qwen2_vl/qwen2_vl.py,sha256=YysRg1PszpZdsH7hywqN2pIMZAbU1P636KZoSJCNri8,5740
|
|
360
360
|
nexaai/mlx_backend/vlm/modeling/models/qwen2_vl/vision.py,sha256=rcneYZYwP5fx3p8xxXEDGvnWAMIn8DXenqBPLBjmVcI,10555
|
|
361
361
|
nexaai/mlx_backend/vlm/modeling/models/qwen3_vl/processor.py,sha256=_WZK8uv0LAGHO8V5LwCojyS4UYk0tfICsZ9GM4_Pfxo,19156
|
|
362
|
-
nexaai/mlx_backend/vlm/modeling/models/qwen3_vl/qwen3vl.py,sha256=
|
|
362
|
+
nexaai/mlx_backend/vlm/modeling/models/qwen3_vl/qwen3vl.py,sha256=rlEJKxPb61bTEBeruqyfrAJtTf_wD0ey-kOmXhgdfVg,50199
|
|
363
363
|
nexaai/mlx_backend/vlm/modeling/models/qwen3_vl/llm_common/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
|
364
364
|
nexaai/mlx_backend/vlm/modeling/models/qwen3_vl/llm_common/base.py,sha256=4RlZwgz8YX2ngmJNaymxFFpw9hJu-0EMw9xwXpngW9o,3496
|
|
365
365
|
nexaai/mlx_backend/vlm/modeling/models/qwen3_vl/llm_common/cache.py,sha256=NMOB6x-RT6svF4H-Ymo5WqnP7ptAal3aaKjWZXWGMsM,17671
|
|
@@ -368,7 +368,7 @@ nexaai/mlx_backend/vlm/modeling/models/qwen3_vl/llm_common/rope_utils.py,sha256=
|
|
|
368
368
|
nexaai/mlx_backend/vlm/modeling/models/qwen3_vl/llm_common/sample_utils.py,sha256=8SEeVwgjuvaYy-4ALAU0RHQMuRr2k7EkXba_csxk498,10673
|
|
369
369
|
nexaai/mlx_backend/vlm/modeling/models/qwen3_vl/llm_common/tokenizer_utils.py,sha256=Gqanx4hBDcon_k5ClhUsS4YpMbZNiee8jvImGS9h43s,13229
|
|
370
370
|
nexaai/mlx_backend/vlm/modeling/models/qwen3vl_moe/processor.py,sha256=_WZK8uv0LAGHO8V5LwCojyS4UYk0tfICsZ9GM4_Pfxo,19156
|
|
371
|
-
nexaai/mlx_backend/vlm/modeling/models/qwen3vl_moe/qwen3vl_moe.py,sha256=
|
|
371
|
+
nexaai/mlx_backend/vlm/modeling/models/qwen3vl_moe/qwen3vl_moe.py,sha256=OrrAQAPY_g5e7Pc8GSL94TXg-CEC8yL3MLDWNOVii-Q,53392
|
|
372
372
|
nexaai/mlx_backend/vlm/modeling/models/qwen3vl_moe/switch_layers.py,sha256=6hjFCvIH8J1UBEPYaIg9HqSakyZgEEiQoJQ9E0WmZoM,5960
|
|
373
373
|
nexaai/mlx_backend/vlm/modeling/models/qwen3vl_moe/llm_common/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
|
374
374
|
nexaai/mlx_backend/vlm/modeling/models/qwen3vl_moe/llm_common/base.py,sha256=4RlZwgz8YX2ngmJNaymxFFpw9hJu-0EMw9xwXpngW9o,3496
|
|
@@ -398,7 +398,7 @@ nexaai/utils/quantization_utils.py,sha256=FYcNSAKGlBqFDUTx3jSKOr2lnq4nyiyC0ZG8oS
|
|
|
398
398
|
nexaai/vlm_impl/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
|
|
399
399
|
nexaai/vlm_impl/mlx_vlm_impl.py,sha256=pLtWm_ckz8a0U-AtAOMVseFDO4OVPvHyYO2KlfBaGYk,10833
|
|
400
400
|
nexaai/vlm_impl/pybind_vlm_impl.py,sha256=FAbhpRJzHgI78r0mUvKybO97R1szvNhH0aTn_I52oT4,8597
|
|
401
|
-
nexaai-1.0.
|
|
402
|
-
nexaai-1.0.
|
|
403
|
-
nexaai-1.0.
|
|
404
|
-
nexaai-1.0.
|
|
401
|
+
nexaai-1.0.19rc9.dist-info/METADATA,sha256=7gh05iEzIQGnTwbB2SedEAYdtSvLXP3m6U8BtkqkuGg,1201
|
|
402
|
+
nexaai-1.0.19rc9.dist-info/WHEEL,sha256=0KYp5feZ1CMUhsfFXKpSQTbSmQbXy4mv6yPPVBXg2EM,110
|
|
403
|
+
nexaai-1.0.19rc9.dist-info/top_level.txt,sha256=LRE2YERlrZk2vfuygnSzsEeqSknnZbz3Z1MHyNmBU4w,7
|
|
404
|
+
nexaai-1.0.19rc9.dist-info/RECORD,,
|
|
File without changes
|
|
File without changes
|