nikto_b/GigaAM-onnx
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

0.1.4.post3

Browse Source
This commit is contained in:
nikto_b
2026-09-16 00:07:29 +03:00
parent 44ee142946
commit 5043855ef8
4 changed files with 21 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+1 -1
View File
@@ -58,7 +58,7 @@ import numpy as np
if 'CUDAExecutionProvider' in ort.get_available_providers(): if 'CUDAExecutionProvider' in ort.get_available_providers():
provider = 'CUDAExecutionProvider' provider = 'CUDAExecutionProvider'
else: else:
provider = 'CPUExecutionProvider" provider = 'CPUExecutionProvider'
opts = ort.SessionOptions() opts = ort.SessionOptions()
opts.intra_op_num_threads = 16 opts.intra_op_num_threads = 16
opts.execution_mode = ort.ExecutionMode.ORT_SEQUENTIAL opts.execution_mode = ort.ExecutionMode.ORT_SEQUENTIAL
pyproject.toml
+12 -8
View File
@@ -4,22 +4,22 @@ authors = [
{ name = "nikto_b", email = "niktob560@yandex.ru" } { name = "nikto_b", email = "niktob560@yandex.ru" }
] ]
license = "MIT" license = "MIT"
version = "0.1.3" version = "0.1.4.post3"
description = "ONNX wrapper for a GigaAMASR models" description = "ONNX wrapper for a GigaAMASR models"
readme = "README.md" readme = "README.md"
requires-python = ">=3.10" requires-python = ">=3.8"
dependencies = [ dependencies = [
"hatchling>=1.28.0", "numpy>=1.24.0",
"numpy==2.*", "onnxruntime>=1.17.0",
"onnxruntime==1.23.*", "sentencepiece>=0.2.0",
"sentencepiece>=0.2.1", "torch>=2.4,<2.9",
"torch>=2.5,<2.9", "torchaudio>=2.4,<2.9",
"torchaudio>=2.5,<2.9",
] ]
[dependency-groups] [dependency-groups]
dev = [ dev = [
"scipy>=1.0.0", "scipy>=1.0.0",
"hatchling",
] ]
[build-system] [build-system]
@@ -28,3 +28,7 @@ build-backend = "hatchling.build"
[tool.hatch.build.targets.wheel] [tool.hatch.build.targets.wheel]
packages = ["src/gigaam_onnx"] packages = ["src/gigaam_onnx"]
core-metadata-version = "2.4"
[tool.hatch.build.targets.sdist]
core-metadata-version = "2.4"
src/gigaam_onnx/asr_abc.py
+3 -2
View File
@@ -76,9 +76,10 @@ class ASRABCModel(abc.ABC):
features = None features = None
input_lengths_array = None input_lengths_array = None
full_batch_size = len(all_wavs) full_batch_size = len(all_wavs)
BATCH_MAX_SIZE = 1
while len(all_wavs) > 0: while len(all_wavs) > 0:
wavs = all_wavs[:128] wavs = all_wavs[:BATCH_MAX_SIZE]
all_wavs = all_wavs[128:] all_wavs = all_wavs[BATCH_MAX_SIZE:]
input_lengths = [] input_lengths = []
processed_wavs = [] processed_wavs = []
src/gigaam_onnx/rnnt.py
+5 -1
View File
@@ -30,6 +30,9 @@ class RNNTASR(ASRABCModel):
timings = [] timings = []
prev_token = self.blank_idx prev_token = self.blank_idx
if len(features.shape) == 2:
features = np.expand_dims(features, axis=0)
pred_states = [ pred_states = [
np.zeros(shape=(1, 1, self.head.decoder.pred_hidden), dtype=DTYPE), np.zeros(shape=(1, 1, self.head.decoder.pred_hidden), dtype=DTYPE),
np.zeros(shape=(1, 1, self.head.decoder.pred_hidden), dtype=DTYPE), np.zeros(shape=(1, 1, self.head.decoder.pred_hidden), dtype=DTYPE),
@@ -57,7 +60,8 @@ class RNNTASR(ASRABCModel):
log_probs = self.jointer.run( log_probs = self.jointer.run(
[node.name for node in self.jointer.get_outputs()], joint_inputs [node.name for node in self.jointer.get_outputs()], joint_inputs
) )
token = log_probs[0].argmax(-1)[0][0] #token = log_probs[0].argmax(-1)[0][0]
token = int(log_probs[0].argmax(-1).reshape(-1)[0])
if token != self.blank_idx: if token != self.blank_idx:
prev_token = int(token) prev_token = int(token)