mertcobanov commited on
Commit
fe29078
·
verified ·
1 Parent(s): eba6803

Ollaya package for knowledgator/gliclass-instruct-large-v1.0

Browse files
README.md ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ base_model:
4
+ - knowledgator/gliclass-instruct-large-v1.0
5
+ library_name: onnx
6
+ tags:
7
+ - ollaya
8
+ - onnx
9
+ - decision-model
10
+ - system-one
11
+ pipeline_tag: text-classification
12
+ ---
13
+
14
+ # gliclass for Ollaya
15
+
16
+ [Ollaya](https://github.com/ollaya-dev/ollaya) package of **[knowledgator/gliclass-instruct-large-v1.0](https://huggingface.co/knowledgator/gliclass-instruct-large-v1.0)** by Knowledgator.
17
+ Ollaya runs open decision models locally, the way Ollama runs LLMs: typed questions in,
18
+ calibrated answers out, behind a TypeSafe-compatible API.
19
+
20
+ ```sh
21
+ ollaya run gliclass
22
+ ```
23
+
24
+ ## What is in this repository
25
+
26
+ This repository holds only the files Ollaya derives, with no weights. Each graph is an ONNX export of the
27
+ original model whose weights **reference the author's own `model.safetensors` by byte offset**,
28
+ so `ollaya pull` downloads the weights from the upstream repository, unmodified and pinned to a
29
+ commit, and verifies their sha256.
30
+
31
+ | Tag | Upstream | Files |
32
+ |---|---|---|
33
+ | `gliclass:large` | [knowledgator/gliclass-instruct-large-v1.0@825e547](https://huggingface.co/knowledgator/gliclass-instruct-large-v1.0/tree/825e5478c1bf4bffbf297690517097ccbdb2e006) | `large/model-fp32.onnx`, `large/decision.json`, `large/calibration.json` |
34
+
35
+ Each tag has an fp32 graph, used on CPU and GPU. Each tag also has `decision.json` (sequence layout, special tokens) and
36
+ `calibration.json` (temperatures).
37
+
38
+ ## Parity
39
+
40
+ Ollaya's Rust runtime matches the Python reference exactly on 482 questions. The token ids and label positions are identical, and so is the decision on every question. Probabilities are within 1e-5, on CPU and CUDA.
41
+
42
+ ## License
43
+
44
+ Same as the upstream model (Apache-2.0). Ollaya itself is Apache-2.0.
large/calibration.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "temperature": [
3
+ 1.0,
4
+ 1.0,
5
+ 1.0
6
+ ],
7
+ "temperature_by_options": {}
8
+ }
large/decision.json ADDED
@@ -0,0 +1,57 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "engine": "onnx",
3
+ "family": "gliclass",
4
+ "layout": "gliclass-uni-v1",
5
+ "contract": "markers",
6
+ "source": {
7
+ "repo": "knowledgator/gliclass-instruct-large-v1.0",
8
+ "revision": "825e5478c1bf4bffbf297690517097ccbdb2e006",
9
+ "reference": "gliclass==0.1.20",
10
+ "license": "Apache-2.0",
11
+ "author": "Knowledgator"
12
+ },
13
+ "encoder": "microsoft/deberta-v3-large",
14
+ "max_len": 1024,
15
+ "tokenize": {
16
+ "add_special_tokens": true,
17
+ "template": "[CLS] $A [SEP]",
18
+ "truncation": "right, to max_len"
19
+ },
20
+ "special_tokens": {
21
+ "cls": 1,
22
+ "sep": 2,
23
+ "pad": 0,
24
+ "label": 128001,
25
+ "label_text": "<<LABEL>>",
26
+ "text_sep": 128002,
27
+ "text_sep_text": "<<SEP>>",
28
+ "example": 128003,
29
+ "example_text": "<<EXAMPLE>>"
30
+ },
31
+ "sanitize": [
32
+ "<<LABEL>>",
33
+ "<<SEP>>",
34
+ "<<EXAMPLE>>"
35
+ ],
36
+ "noul": {
37
+ "pair_when": "both true and false descriptions are non-empty",
38
+ "pair_labels": [
39
+ "false",
40
+ "true"
41
+ ],
42
+ "single_label": "instructions",
43
+ "single_prompt": ""
44
+ },
45
+ "inputs": [
46
+ "input_ids",
47
+ "attention_mask",
48
+ "marker_pos",
49
+ "marker_mask",
50
+ "qtype"
51
+ ],
52
+ "outputs": [
53
+ "logits"
54
+ ],
55
+ "min_markers": 1,
56
+ "opset": 20
57
+ }
large/model-fp32.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4e3b560979faf451ee3cdfd9e9f6e0e32efbd4e50992ef89fd085a6fffdd27a5
3
+ size 5065943