offlinedemo 0.3.0__tar.gz → 0.5.0__tar.gz

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.
Files changed (8) hide show
  1. offlinedemo-0.5.0/PKG-INFO +56 -0
  2. {offlinedemo-0.3.0 → offlinedemo-0.5.0}/README.md +7 -2
  3. offlinedemo-0.5.0/pyproject.toml +22 -0
  4. offlinedemo-0.5.0/pyproject.toml.orig +21 -0
  5. {offlinedemo-0.3.0 → offlinedemo-0.5.0}/src/offlinedemo/__init__.py +8 -6
  6. offlinedemo-0.3.0/PKG-INFO +0 -106
  7. offlinedemo-0.3.0/pyproject.toml +0 -77
  8. offlinedemo-0.3.0/pyproject.toml.orig +0 -76
offlinedemo-0.5.0/PKG-INFO ADDED
@@ -0,0 +1,56 @@
1
+ Metadata-Version: 2.3
2
+ Name: offlinedemo
3
+ Version: 0.5.0
4
+ Summary: Easy demonstration of models by offlineisbetter.
5
+ Author: offlineisbetter
6
+ Author-email: offlineisbetter <dev@offlineisbetter.com>
7
+ Requires-Dist: numpy>=2.5.3
8
+ Requires-Dist: onnxruntime>=1.30.0
9
+ Requires-Dist: tokenizers>=0.23.2
10
+ Requires-Python: >=3.13
11
+ Description-Content-Type: text/markdown
12
+
13
+ # _offlineisbetter_
14
+
15
+ inference for models by _offlineisbetter_.
16
+
17
+ we believe that you shouldn't give your data to faceless companies, that you deserve to run text models locally, and that you shouldn't need to buy expensive hardware. so we're building _offlineisbetter_.
18
+
19
+ _offlineisbetter_ models will be for _encoding_ tasks: sentiment analysis, text tagging, document retrieval, etc., rather than for _decoding_ tasks like autoregressive generation. we believe that it's wasteful and dangerous to depend on cloud apis for frontier language models to do these simple tasks, and it should be almost mindless to download a model to _use_ it without dealing with runtimes or quantization formats.
20
+
21
+ ## try it yourself
22
+
23
+ try our first model yourself. our first model is a small (230m) text model for sentiment analysis called `offline-sentiment-small`.
24
+
25
+ ```bash
26
+ pip install offlinedemo
27
+ ```
28
+
29
+ download the model archive from the releases page of this repo and unpack the model.
30
+
31
+ ```bash
32
+ tar -xvf offline-sentiment-small.tar
33
+ ```
34
+
35
+ run the demo, passing the inflated directory containing the model checkpoint.
36
+
37
+ ```bash
38
+ offlinedemo offline-sentiment-small
39
+ ```
40
+
41
+ ## benchmarks
42
+
43
+ below are benchmarks for `offline-sentiment-small` on binary sentiment classification using the [stanfordnlp/sst2](https://huggingface.co/datasets/stanfordnlp/sst2) validation set. all benchmarks were completed on the ryzen 9950x3d cpu on one thread.
44
+
45
+ critically: _we only allowed 5 minutes of setup time_ for each model. our focus is on improving the developer's experience and reducing headache! that means that `*bert` models were used in their original form (`torch`), as downloaded from hugging face, as was ours (`onnx`). this is _not an apples-to-apples comparison_ in the same runtime, and that's intentional.
46
+
47
+ | model | parameters | p95 (ms) | f1 (validation) |
48
+ |:---:|:---:|:---:|:---:|
49
+ | `offline-sentiment-small` | 230m | 80.32 | 0.9489 |
50
+ | `distilbert-base` | 67m | 66.15 | 0.9321 |
51
+ | `roberta-base` | 125m | 469.65 | 0.9396 |
52
+ | `modernbert-base` | 149m | 530.73 | 0.9396 |
53
+
54
+ ## philosophy
55
+
56
+ succinctly, the core philosophy of _offlineisbetter_ is that parameter-efficient and low-latency models should be easily accessible to everybody. of course hugging face and `transformers.pipeline` allows you to run sentiment analysis in three lines of python, but for more parameter-efficient models, already quantized and with optimized computation graphs.
{offlinedemo-0.3.0 → offlinedemo-0.5.0}/README.md RENAMED
@@ -14,7 +14,7 @@ try our first model yourself. our first model is a small (230m) text model for s
14
14
  pip install offlinedemo
15
15
  ```
16
16
 
17
- download the model archive from the website and unpack the model.
17
+ download the model archive from the releases page of this repo and unpack the model.
18
18
 
19
19
  ```bash
20
20
  tar -xvf offline-sentiment-small.tar
@@ -28,11 +28,16 @@ offlinedemo offline-sentiment-small
28
28
 
29
29
  ## benchmarks
30
30
 
31
+ below are benchmarks for `offline-sentiment-small` on binary sentiment classification using the [stanfordnlp/sst2](https://huggingface.co/datasets/stanfordnlp/sst2) validation set. all benchmarks were completed on the ryzen 9950x3d cpu on one thread.
32
+
33
+ critically: _we only allowed 5 minutes of setup time_ for each model. our focus is on improving the developer's experience and reducing headache! that means that `*bert` models were used in their original form (`torch`), as downloaded from hugging face, as was ours (`onnx`). this is _not an apples-to-apples comparison_ in the same runtime, and that's intentional.
34
+
31
35
  | model | parameters | p95 (ms) | f1 (validation) |
32
36
  |:---:|:---:|:---:|:---:|
33
37
  | `offline-sentiment-small` | 230m | 80.32 | 0.9489 |
34
38
  | `distilbert-base` | 67m | 66.15 | 0.9321 |
35
- | `roberta-base` |
39
+ | `roberta-base` | 125m | 469.65 | 0.9396 |
40
+ | `modernbert-base` | 149m | 530.73 | 0.9396 |
36
41
 
37
42
  ## philosophy
38
43
 
offlinedemo-0.5.0/pyproject.toml ADDED
@@ -0,0 +1,22 @@
1
+ [project]
2
+ name = "offlinedemo"
3
+ version = "0.5.0"
4
+ description = "Easy demonstration of models by offlineisbetter."
5
+ readme = "README.md"
6
+ requires-python = ">=3.13"
7
+ dependencies = [
8
+ "numpy>=2.5.3",
9
+ "onnxruntime>=1.30.0",
10
+ "tokenizers>=0.23.2",
11
+ ]
12
+
13
+ [[project.authors]]
14
+ name = "offlineisbetter"
15
+ email = "dev@offlineisbetter.com"
16
+
17
+ [project.scripts]
18
+ offlinedemo = "offlinedemo:main"
19
+
20
+ [build-system]
21
+ requires = ["uv_build>=0.12.19,<0.13.0"]
22
+ build-backend = "uv_build"
offlinedemo-0.5.0/pyproject.toml.orig ADDED
@@ -0,0 +1,21 @@
1
+ [project]
2
+ name = "offlinedemo"
3
+ version = "0.5.0"
4
+ description = "Easy demonstration of models by offlineisbetter."
5
+ readme = "README.md"
6
+ authors = [
7
+ { name = "offlineisbetter", email = "dev@offlineisbetter.com" }
8
+ ]
9
+ requires-python = ">=3.13"
10
+ dependencies = [
11
+ "numpy>=2.5.3",
12
+ "onnxruntime>=1.30.0",
13
+ "tokenizers>=0.23.2",
14
+ ]
15
+
16
+ [project.scripts]
17
+ offlinedemo = "offlinedemo:main"
18
+
19
+ [build-system]
20
+ requires = ["uv_build>=0.12.19,<0.13.0"]
21
+ build-backend = "uv_build"
{offlinedemo-0.3.0 → offlinedemo-0.5.0}/src/offlinedemo/__init__.py RENAMED
@@ -3,11 +3,12 @@
3
3
 
4
4
  from pathlib import Path
5
5
  import json
6
+ import os
6
7
  import sys
7
8
  import time
8
9
 
9
10
  import numpy as np
10
- from transformers import AutoTokenizer
11
+ from tokenizers import Tokenizer
11
12
  import onnxruntime as ort
12
13
 
13
14
  def main():
@@ -30,10 +31,7 @@ def main():
30
31
  classes = {v: k for k, v in classes.items()}
31
32
 
32
33
  # Load tokenizer
33
- tokenizer = AutoTokenizer.from_pretrained(
34
- checkpoint,
35
- local_files_only = True,
36
- )
34
+ tokenizer = Tokenizer.from_file(str(checkpoint / "tokenizer.json"))
37
35
  print("Tokenizer loaded!")
38
36
  print()
39
37
 
@@ -42,7 +40,11 @@ def main():
42
40
 
43
41
  # Tokenize and run inference
44
42
  start = time.perf_counter()
45
- tokens = tokenizer(user_input)
43
+ encoded = tokenizer.encode(user_input)
44
+ tokens = {
45
+ "input_ids": np.array(encoded.ids),
46
+ "attention_mask": np.array(encoded.attention_mask),
47
+ }
46
48
  result = session.run(None, tokens)[0][0]
47
49
  idx = int(np.argmax(result))
48
50
  c = classes[idx]
offlinedemo-0.3.0/PKG-INFO DELETED
@@ -1,106 +0,0 @@
1
- Metadata-Version: 2.3
2
- Name: offlinedemo
3
- Version: 0.3.0
4
- Summary: Easy demonstration of models by offlineisbetter.
5
- Author: offlineisbetter
6
- Author-email: offlineisbetter <dev@offlineisbetter.com>
7
- Requires-Dist: annotated-doc==0.0.5
8
- Requires-Dist: anyio==4.15.1
9
- Requires-Dist: certifi==2026.7.22
10
- Requires-Dist: click==8.5.0
11
- Requires-Dist: cuda-bindings==13.4.3
12
- Requires-Dist: cuda-pathfinder==1.8.2
13
- Requires-Dist: cuda-toolkit==13.0.3.0
14
- Requires-Dist: filelock==4.0.6
15
- Requires-Dist: flatbuffers==25.12.19
16
- Requires-Dist: fsspec==2026.9.0
17
- Requires-Dist: h11==0.16.0
18
- Requires-Dist: hf-xet==1.6.0
19
- Requires-Dist: httpcore==1.0.9
20
- Requires-Dist: httpx==0.28.1
21
- Requires-Dist: huggingface-hub==1.33.0
22
- Requires-Dist: idna==3.20
23
- Requires-Dist: jinja2==3.1.6
24
- Requires-Dist: markdown-it-py==4.2.0
25
- Requires-Dist: markupsafe==3.0.3
26
- Requires-Dist: mdurl==0.1.2
27
- Requires-Dist: ml-dtypes==0.6.0
28
- Requires-Dist: mpmath==1.3.0
29
- Requires-Dist: networkx==3.7
30
- Requires-Dist: numpy==2.5.3
31
- Requires-Dist: nvidia-cublas==13.1.1.3
32
- Requires-Dist: nvidia-cuda-cupti==13.0.85
33
- Requires-Dist: nvidia-cuda-nvrtc==13.0.88
34
- Requires-Dist: nvidia-cuda-runtime==13.0.96
35
- Requires-Dist: nvidia-cudnn-cu13==9.24.0.43
36
- Requires-Dist: nvidia-cufft==12.0.0.61
37
- Requires-Dist: nvidia-cufile==1.15.1.6
38
- Requires-Dist: nvidia-curand==10.4.0.35
39
- Requires-Dist: nvidia-cusolver==12.0.4.66
40
- Requires-Dist: nvidia-cusparse==12.6.3.3
41
- Requires-Dist: nvidia-cusparselt-cu13==0.8.1
42
- Requires-Dist: nvidia-nccl-cu13==2.30.7
43
- Requires-Dist: nvidia-nvjitlink==13.4.92
44
- Requires-Dist: nvidia-nvshmem-cu13==3.4.5
45
- Requires-Dist: nvidia-nvtx==13.0.85
46
- Requires-Dist: onnx==1.23.1
47
- Requires-Dist: onnxruntime==1.30.0
48
- Requires-Dist: packaging==26.3
49
- Requires-Dist: protobuf==7.36.2
50
- Requires-Dist: pygments==2.21.0
51
- Requires-Dist: pyyaml==6.0.3
52
- Requires-Dist: regex==2026.9.29
53
- Requires-Dist: rich==15.0.0
54
- Requires-Dist: safetensors==0.8.0
55
- Requires-Dist: setuptools==84.0.0
56
- Requires-Dist: shellingham==1.5.4
57
- Requires-Dist: sympy==1.14.0
58
- Requires-Dist: tokenizers==0.23.2
59
- Requires-Dist: torch==2.14.0
60
- Requires-Dist: tqdm==4.70.1
61
- Requires-Dist: transformers==5.17.0
62
- Requires-Dist: triton==3.8.0
63
- Requires-Dist: typer==0.27.2
64
- Requires-Dist: typing-extensions==4.16.0
65
- Requires-Python: >=3.13
66
- Description-Content-Type: text/markdown
67
-
68
- # _offlineisbetter_
69
-
70
- inference for models by _offlineisbetter_.
71
-
72
- we believe that you shouldn't give your data to faceless companies, that you deserve to run text models locally, and that you shouldn't need to buy expensive hardware. so we're building _offlineisbetter_.
73
-
74
- _offlineisbetter_ models will be for _encoding_ tasks: sentiment analysis, text tagging, document retrieval, etc., rather than for _decoding_ tasks like autoregressive generation. we believe that it's wasteful and dangerous to depend on cloud apis for frontier language models to do these simple tasks, and it should be almost mindless to download a model to _use_ it without dealing with runtimes or quantization formats.
75
-
76
- ## try it yourself
77
-
78
- try our first model yourself. our first model is a small (230m) text model for sentiment analysis called `offline-sentiment-small`.
79
-
80
- ```bash
81
- pip install offlinedemo
82
- ```
83
-
84
- download the model archive from the website and unpack the model.
85
-
86
- ```bash
87
- tar -xvf offline-sentiment-small.tar
88
- ```
89
-
90
- run the demo, passing the inflated directory containing the model checkpoint.
91
-
92
- ```bash
93
- offlinedemo offline-sentiment-small
94
- ```
95
-
96
- ## benchmarks
97
-
98
- | model | parameters | p95 (ms) | f1 (validation) |
99
- |:---:|:---:|:---:|:---:|
100
- | `offline-sentiment-small` | 230m | 80.32 | 0.9489 |
101
- | `distilbert-base` | 67m | 66.15 | 0.9321 |
102
- | `roberta-base` |
103
-
104
- ## philosophy
105
-
106
- succinctly, the core philosophy of _offlineisbetter_ is that parameter-efficient and low-latency models should be easily accessible to everybody. of course hugging face and `transformers.pipeline` allows you to run sentiment analysis in three lines of python, but for more parameter-efficient models, already quantized and with optimized computation graphs.
offlinedemo-0.3.0/pyproject.toml DELETED
@@ -1,77 +0,0 @@
1
- [project]
2
- name = "offlinedemo"
3
- version = "0.3.0"
4
- description = "Easy demonstration of models by offlineisbetter."
5
- readme = "README.md"
6
- requires-python = ">=3.13"
7
- dependencies = [
8
- "annotated-doc==0.0.5",
9
- "anyio==4.15.1",
10
- "certifi==2026.7.22",
11
- "click==8.5.0",
12
- "cuda-bindings==13.4.3",
13
- "cuda-pathfinder==1.8.2",
14
- "cuda-toolkit==13.0.3.0",
15
- "filelock==4.0.6",
16
- "flatbuffers==25.12.19",
17
- "fsspec==2026.9.0",
18
- "h11==0.16.0",
19
- "hf-xet==1.6.0",
20
- "httpcore==1.0.9",
21
- "httpx==0.28.1",
22
- "huggingface-hub==1.33.0",
23
- "idna==3.20",
24
- "jinja2==3.1.6",
25
- "markdown-it-py==4.2.0",
26
- "markupsafe==3.0.3",
27
- "mdurl==0.1.2",
28
- "ml-dtypes==0.6.0",
29
- "mpmath==1.3.0",
30
- "networkx==3.7",
31
- "numpy==2.5.3",
32
- "nvidia-cublas==13.1.1.3",
33
- "nvidia-cuda-cupti==13.0.85",
34
- "nvidia-cuda-nvrtc==13.0.88",
35
- "nvidia-cuda-runtime==13.0.96",
36
- "nvidia-cudnn-cu13==9.24.0.43",
37
- "nvidia-cufft==12.0.0.61",
38
- "nvidia-cufile==1.15.1.6",
39
- "nvidia-curand==10.4.0.35",
40
- "nvidia-cusolver==12.0.4.66",
41
- "nvidia-cusparse==12.6.3.3",
42
- "nvidia-cusparselt-cu13==0.8.1",
43
- "nvidia-nccl-cu13==2.30.7",
44
- "nvidia-nvjitlink==13.4.92",
45
- "nvidia-nvshmem-cu13==3.4.5",
46
- "nvidia-nvtx==13.0.85",
47
- "onnx==1.23.1",
48
- "onnxruntime==1.30.0",
49
- "packaging==26.3",
50
- "protobuf==7.36.2",
51
- "pygments==2.21.0",
52
- "pyyaml==6.0.3",
53
- "regex==2026.9.29",
54
- "rich==15.0.0",
55
- "safetensors==0.8.0",
56
- "setuptools==84.0.0",
57
- "shellingham==1.5.4",
58
- "sympy==1.14.0",
59
- "tokenizers==0.23.2",
60
- "torch==2.14.0",
61
- "tqdm==4.70.1",
62
- "transformers==5.17.0",
63
- "triton==3.8.0",
64
- "typer==0.27.2",
65
- "typing-extensions==4.16.0",
66
- ]
67
-
68
- [[project.authors]]
69
- name = "offlineisbetter"
70
- email = "dev@offlineisbetter.com"
71
-
72
- [project.scripts]
73
- offlinedemo = "offlinedemo:main"
74
-
75
- [build-system]
76
- requires = ["uv_build>=0.12.19,<0.13.0"]
77
- build-backend = "uv_build"
offlinedemo-0.3.0/pyproject.toml.orig DELETED
@@ -1,76 +0,0 @@
1
- [project]
2
- name = "offlinedemo"
3
- version = "0.3.0"
4
- description = "Easy demonstration of models by offlineisbetter."
5
- readme = "README.md"
6
- authors = [
7
- { name = "offlineisbetter", email = "dev@offlineisbetter.com" }
8
- ]
9
- requires-python = ">=3.13"
10
- dependencies = [
11
- "annotated-doc==0.0.5",
12
- "anyio==4.15.1",
13
- "certifi==2026.7.22",
14
- "click==8.5.0",
15
- "cuda-bindings==13.4.3",
16
- "cuda-pathfinder==1.8.2",
17
- "cuda-toolkit==13.0.3.0",
18
- "filelock==4.0.6",
19
- "flatbuffers==25.12.19",
20
- "fsspec==2026.9.0",
21
- "h11==0.16.0",
22
- "hf-xet==1.6.0",
23
- "httpcore==1.0.9",
24
- "httpx==0.28.1",
25
- "huggingface-hub==1.33.0",
26
- "idna==3.20",
27
- "jinja2==3.1.6",
28
- "markdown-it-py==4.2.0",
29
- "markupsafe==3.0.3",
30
- "mdurl==0.1.2",
31
- "ml-dtypes==0.6.0",
32
- "mpmath==1.3.0",
33
- "networkx==3.7",
34
- "numpy==2.5.3",
35
- "nvidia-cublas==13.1.1.3",
36
- "nvidia-cuda-cupti==13.0.85",
37
- "nvidia-cuda-nvrtc==13.0.88",
38
- "nvidia-cuda-runtime==13.0.96",
39
- "nvidia-cudnn-cu13==9.24.0.43",
40
- "nvidia-cufft==12.0.0.61",
41
- "nvidia-cufile==1.15.1.6",
42
- "nvidia-curand==10.4.0.35",
43
- "nvidia-cusolver==12.0.4.66",
44
- "nvidia-cusparse==12.6.3.3",
45
- "nvidia-cusparselt-cu13==0.8.1",
46
- "nvidia-nccl-cu13==2.30.7",
47
- "nvidia-nvjitlink==13.4.92",
48
- "nvidia-nvshmem-cu13==3.4.5",
49
- "nvidia-nvtx==13.0.85",
50
- "onnx==1.23.1",
51
- "onnxruntime==1.30.0",
52
- "packaging==26.3",
53
- "protobuf==7.36.2",
54
- "pygments==2.21.0",
55
- "pyyaml==6.0.3",
56
- "regex==2026.9.29",
57
- "rich==15.0.0",
58
- "safetensors==0.8.0",
59
- "setuptools==84.0.0",
60
- "shellingham==1.5.4",
61
- "sympy==1.14.0",
62
- "tokenizers==0.23.2",
63
- "torch==2.14.0",
64
- "tqdm==4.70.1",
65
- "transformers==5.17.0",
66
- "triton==3.8.0",
67
- "typer==0.27.2",
68
- "typing-extensions==4.16.0",
69
- ]
70
-
71
- [project.scripts]
72
- offlinedemo = "offlinedemo:main"
73
-
74
- [build-system]
75
- requires = ["uv_build>=0.12.19,<0.13.0"]
76
- build-backend = "uv_build"