HashNuke commited on
Commit
dcbadfd
·
verified ·
1 Parent(s): 71fea4d

Add the first experimental e5-small checkpoint

Browse files

Preserve epoch-5 weights and tokenizer, original source hashes and training metrics. Model card records synthetic-test limitations and short-interruption failures.

.gitattributes CHANGED
@@ -1,35 +1,3 @@
1
- *.7z filter=lfs diff=lfs merge=lfs -text
2
- *.arrow filter=lfs diff=lfs merge=lfs -text
3
- *.bin filter=lfs diff=lfs merge=lfs -text
4
- *.bz2 filter=lfs diff=lfs merge=lfs -text
5
- *.ckpt filter=lfs diff=lfs merge=lfs -text
6
- *.ftz filter=lfs diff=lfs merge=lfs -text
7
- *.gz filter=lfs diff=lfs merge=lfs -text
8
- *.h5 filter=lfs diff=lfs merge=lfs -text
9
- *.joblib filter=lfs diff=lfs merge=lfs -text
10
- *.lfs.* filter=lfs diff=lfs merge=lfs -text
11
- *.mlmodel filter=lfs diff=lfs merge=lfs -text
12
- *.model filter=lfs diff=lfs merge=lfs -text
13
- *.msgpack filter=lfs diff=lfs merge=lfs -text
14
- *.npy filter=lfs diff=lfs merge=lfs -text
15
- *.npz filter=lfs diff=lfs merge=lfs -text
16
- *.onnx filter=lfs diff=lfs merge=lfs -text
17
- *.ot filter=lfs diff=lfs merge=lfs -text
18
- *.parquet filter=lfs diff=lfs merge=lfs -text
19
- *.pb filter=lfs diff=lfs merge=lfs -text
20
- *.pickle filter=lfs diff=lfs merge=lfs -text
21
- *.pkl filter=lfs diff=lfs merge=lfs -text
22
- *.pt filter=lfs diff=lfs merge=lfs -text
23
- *.pth filter=lfs diff=lfs merge=lfs -text
24
- *.rar filter=lfs diff=lfs merge=lfs -text
25
  *.safetensors filter=lfs diff=lfs merge=lfs -text
26
- saved_model/**/* filter=lfs diff=lfs merge=lfs -text
27
- *.tar.* filter=lfs diff=lfs merge=lfs -text
28
- *.tar filter=lfs diff=lfs merge=lfs -text
29
- *.tflite filter=lfs diff=lfs merge=lfs -text
30
- *.tgz filter=lfs diff=lfs merge=lfs -text
31
- *.wasm filter=lfs diff=lfs merge=lfs -text
32
- *.xz filter=lfs diff=lfs merge=lfs -text
33
- *.zip filter=lfs diff=lfs merge=lfs -text
34
- *.zst filter=lfs diff=lfs merge=lfs -text
35
- *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  *.safetensors filter=lfs diff=lfs merge=lfs -text
2
+ /tokenizer.json filter=lfs diff=lfs merge=lfs -text
3
+ tokenizer.json filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
LICENSE ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ MIT License
2
+
3
+ Copyright (c) 2026 Akash Manohar
4
+
5
+ Permission is hereby granted, free of charge, to any person obtaining a copy
6
+ of this software and associated documentation files (the "Software"), to deal
7
+ in the Software without restriction, including without limitation the rights
8
+ to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
9
+ copies of the Software, and to permit persons to whom the Software is
10
+ furnished to do so, subject to the following conditions:
11
+
12
+ The above copyright notice and this permission notice shall be included in all
13
+ copies or substantial portions of the Software.
14
+
15
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16
+ IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17
+ FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
18
+ AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19
+ LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20
+ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21
+ SOFTWARE.
NOTICE ADDED
@@ -0,0 +1 @@
 
 
1
+ This model is fine-tuned from intfloat/multilingual-e5-small, distributed under the MIT licence. Base model authors and documentation: https://huggingface.co/intfloat/multilingual-e5-small
README.md ADDED
@@ -0,0 +1,115 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language: [en, hi, kn, ta, te, ml, zh, ja, id, fil, fr, de, es, it]
3
+ license: mit
4
+ library_name: transformers
5
+ pipeline_tag: text-classification
6
+ base_model: intfloat/multilingual-e5-small
7
+ datasets: [HashNuke/barge-in-detector-data]
8
+ tags: [barge-in, backchannel, multilingual, experimental]
9
+ ---
10
+ # Barge-in detector: experimental checkpoint
11
+
12
+ Classifies a caller's cleaned utterance spoken **while a voice agent is talking**.
13
+ `backchannel` lets it continue; `barge_in` yields the floor. This repo owns
14
+ weights, tokenizer, training, inference, evaluation and benchmarks.
15
+ [The dataset repo](https://huggingface.co/datasets/HashNuke/barge-in-detector-data)
16
+ owns generation, cleaning, splits, review and historical labnotes.
17
+
18
+ **Not production-ready:** it misses 25/30 short English challenge interruptions,
19
+ including “stop” and “wait”. Probabilities are not calibrated. The runtime owns
20
+ thresholds and handles answers after the agent finishes speaking.
21
+
22
+ ## Checkpoint and evidence
23
+
24
+ Root model.safetensors, config.json, tokenizer.json and tokenizer_config.json
25
+ are the selected epoch-5 fp32 checkpoint. Base: intfloat/multilingual-e5-small
26
+ (MIT); see LICENSE and NOTICE. `release.json` identifies dataset/artifact hashes
27
+ and the published dataset revision. `reports/source-manifest.json` captures the
28
+ original source snapshot, which predates the repository separation.
29
+
30
+ Training: 2,640 synthetic rows; 333 validation; 331 test; five epochs, batch 32,
31
+ weighted cross-entropy, AdamW, learning rate 3e-5, 10% warm-up, seed 20261115.
32
+ BF16 on one RTX 5090. Epoch selected using validation macro F1 (0.99324).
33
+ Full training history and breakdowns are in reports/training-metrics.json.
34
+
35
+ | Evaluation | Accuracy | Missed interruptions | False interruptions |
36
+ | --- | --- | --- | --- |
37
+ | Synthetic validation (333) | 99.4% | 0.45% | 0.90% |
38
+ | Synthetic test (331) | 100% | 0% | 0% |
39
+ | Reviewed English challenge (98) | 73.5% | 54.2% | 0% |
40
+ | Challenge unseen in any synthetic split (93) | 72.0% | 54.2% | 0% |
41
+
42
+ The challenge was authored before inference, owner-reviewed and reconciled with
43
+ authorization. Final binary review labels match the frozen authored labels.
44
+ It is deliberately adversarial, includes repeated prefixes, and is now a
45
+ regression/development set. Synthetic splits share generation criteria; exact
46
+ cleaned-text leakage is absent, but template and semantic leakage remain unaudited.
47
+ These scores do not establish representative real-world accuracy. All 14 listed
48
+ languages have synthetic results; human-verified accuracy across all languages
49
+ has not been established.
50
+
51
+ ## Infer on CPU
52
+
53
+ ```shell
54
+ uv sync
55
+ uv run python -m barge_in_detector_model.infer "please stop for a moment"
56
+ ```
57
+
58
+ Input must already follow the dataset cleaner: lowercase, punctuation removed,
59
+ native scripts optionally mixed with English. No agent context, audio or language
60
+ hint is needed. Keep the first 64 tokenizer tokens. Labels are 0=backchannel,
61
+ 1=barge_in. The command emits `p_barge_in` and a label at the 0.5 threshold.
62
+
63
+ ```python
64
+ import torch
65
+ from transformers import AutoTokenizer, AutoModelForSequenceClassification
66
+
67
+ repo = "HashNuke/barge-in-detector"
68
+ tokenizer = AutoTokenizer.from_pretrained(repo)
69
+ model = AutoModelForSequenceClassification.from_pretrained(repo).cpu().eval()
70
+ with torch.inference_mode():
71
+ enc = tokenizer("please stop for a moment", return_tensors="pt", truncation=True, max_length=64)
72
+ p_barge_in = model(**enc).logits.softmax(-1)[0, 1].item()
73
+ ```
74
+
75
+ ## Train and evaluate
76
+
77
+ ```shell
78
+ uv sync --group dev
79
+ uv run python -m barge_in_detector_model.train --out models/run2 --dataset-revision DATASET_COMMIT
80
+ uv run python -m barge_in_detector_model.evaluate --model models/run2 --split test --dataset-revision DATASET_COMMIT
81
+ ```
82
+
83
+ Data is fetched from the separate dataset repository at a resolved immutable
84
+ revision. `--splits /path/to/dataset/data/splits` supports an existing checkout;
85
+ retain its revision and file hashes with the run. Tune settings and thresholds
86
+ on validation, report test without tuning, and never train on challenge rows.
87
+ Migration does not retrain this checkpoint. Read AGENTS.md and the model
88
+ milestone index before starting work. Use `bin/create-labnotes` for checkpoints.
89
+
90
+ ## CPU latency
91
+
92
+ Local four-vCPU AMD EPYC Genoa VM; eager PyTorch 2.15.0+cpu fp32; four threads,
93
+ batch one, 100 warm-ups and 1,000 measured calls per scenario:
94
+
95
+ | Warm text through label | p50 | p95 | p99 |
96
+ | --- | --- | --- | --- |
97
+ | Typical multilingual test input | 13.1 ms | 19.4 ms | 25.3 ms |
98
+ | Full 64-token budget | 26.8 ms | 33.3 ms | 37.9 ms |
99
+
100
+ Cold Python process through first label: median 4.22 seconds, range 3.85–4.81
101
+ seconds across ten runs. Includes imports and local model/tokenizer load; OS file
102
+ cache may be warm. Peak warm process RSS approximately 1,164 MiB. Network, HTTP,
103
+ queueing and Vxpipe boundary are excluded. Reference-CPU gates are unestablished;
104
+ ONNX/int8 is unmeasured. Reports retain the original benchmark provenance.
105
+
106
+ ```shell
107
+ uv run python -m experiments.benchmark_cpu --threads 4 --splits /path/to/dataset/data/splits
108
+ uv run python -m experiments.benchmark_cpu_cold --threads 4
109
+ uv run python -m experiments.evaluate_english_challenge --dataset-dir /path/to/dataset
110
+ ```
111
+
112
+ For CPU-only use, install the matching CPU PyTorch wheel from the official CPU
113
+ index. The standard lock also covers GPU training dependencies. Do CPU tests
114
+ locally; rent only a single datacenter RTX 5090 below $0.65/hour for training,
115
+ using the Vast.ai PyTorch image and tmux session gpubox.
config.json ADDED
@@ -0,0 +1,38 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_cross_attention": false,
3
+ "architectures": [
4
+ "BertForSequenceClassification"
5
+ ],
6
+ "attention_probs_dropout_prob": 0.1,
7
+ "bos_token_id": null,
8
+ "classifier_dropout": null,
9
+ "dtype": "float32",
10
+ "eos_token_id": null,
11
+ "hidden_act": "gelu",
12
+ "hidden_dropout_prob": 0.1,
13
+ "hidden_size": 384,
14
+ "id2label": {
15
+ "0": "backchannel",
16
+ "1": "barge_in"
17
+ },
18
+ "initializer_range": 0.02,
19
+ "intermediate_size": 1536,
20
+ "is_decoder": false,
21
+ "label2id": {
22
+ "backchannel": 0,
23
+ "barge_in": 1
24
+ },
25
+ "layer_norm_eps": 1e-12,
26
+ "max_position_embeddings": 512,
27
+ "model_type": "bert",
28
+ "num_attention_heads": 12,
29
+ "num_hidden_layers": 12,
30
+ "pad_token_id": 0,
31
+ "position_embedding_type": "absolute",
32
+ "tie_word_embeddings": true,
33
+ "tokenizer_class": "XLMRobertaTokenizer",
34
+ "transformers_version": "5.23.0",
35
+ "type_vocab_size": 2,
36
+ "use_cache": true,
37
+ "vocab_size": 250037
38
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bddb65c08f6260cfc9d320e65b4229229ec9f3ca95204aed98ab5ae45ff758d3
3
+ size 470641664
reports/source-manifest.json ADDED
@@ -0,0 +1,73 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ ".gitattributes": "d76d90c6cfbe7c1e793b13dc70485f1a71c80b19c72e7f3a2829786ac9eef810",
3
+ ".gitignore": "01e2dc4c57a1279bf408ae19a88c62b2e29e92ded5bdc337ca4cadff47dd92fc",
4
+ ".python-version": "a876e0b10411037a012498b9fe18d9bc1df32ed8b722a13564dc944ddcfd9135",
5
+ "AGENTS.md": "e9e62c3227ce52f8ba930c3f5119816c168a7f1477f102eb7fd781598c68b1f1",
6
+ "barge_in_detector/__init__.py": "28de6fe1d126dec7b9cd85273f805831b24fb7c443ed2707c47552caede29072",
7
+ "barge_in_detector/__main__.py": "8efaf592230270b84d1ba8513bdb244b104aba73a517f0734197ccbfa18afb8a",
8
+ "barge_in_detector/catalogue.py": "8b59351941bce2a92420ea0b13989a4c002a52d23c6279c10331645fac9796e2",
9
+ "barge_in_detector/cleaner.py": "58bcb1c3f21f66aecb534cbc20e07d5350c3d8d4c92e2084c4943ca9493c3a22",
10
+ "barge_in_detector/cli.py": "a68d6d7348ab34520f0d8ba88731367da6af0fb574548a290e43f45bd8941517",
11
+ "barge_in_detector/generation/__init__.py": "b81aff66375501799977abd59786a5ad9da99f6998febc5fac64a016a6cea3fd",
12
+ "barge_in_detector/generation/config.py": "844d5945b93c44ff04a35f6f161a1ab29a29d39ffacc7af1b268276818d20662",
13
+ "barge_in_detector/generation/llm.py": "972ee568321255f6ceeee3f34c323eedac5b5ec83062d67cc662b53eee86b7a9",
14
+ "barge_in_detector/generation/prompts.py": "04c3b61b4ef0a7cbd5edcb0438f544b564c602e0d213291dafbcff9fe1f086ce",
15
+ "barge_in_detector/generation/responses.py": "cb5849ddbc90e800bd8c20a17896c0f7cb406b6eacf7126ddc2a1d6468c1ef5d",
16
+ "barge_in_detector/generation/utterance_criteria.py": "78b938e758fa5f8d8d6dedbc48096747a8521eca6812822ef8cccfd70ff62c72",
17
+ "barge_in_detector/generation/utterance_runner.py": "51d383bee865e79242997a6ef1a3ac2add896fb86969689cdbd7a8d26cf734a4",
18
+ "barge_in_detector/generation/utterances.py": "535f7c54e9dff1ed61903713b8c825e98b453932b753da61b46d41479374cd42",
19
+ "barge_in_detector/ids.py": "7a5abdf7ebfb64e5fb1588f887fd951a0df345d71c90d59b445ef2a8c1bf214e",
20
+ "barge_in_detector/storage/__init__.py": "4e44293470af3982b64c05c2d17a949185c10b4bd53bf72b5ba6229f51507db4",
21
+ "barge_in_detector/storage/schema.py": "0ccc76ec546a0fe1c414fed1f9f866b0103a55319f04e6d5d3bed12e83862e03",
22
+ "barge_in_detector/storage/store.py": "b3b32e195dc06653702e307d7199cc380b4a74dd687ab860836d491c8166311f",
23
+ "barge_in_detector/training/__init__.py": "230e5ab5c3a00f056980f02d06acc65a0efe31b7738a18caf9c7c6de1b33c1d1",
24
+ "barge_in_detector/training/evaluate.py": "4762ae1a3df0b942049b1eb4dd9faad7e37b5c77bb8200f2041e9e3f30ee1e5e",
25
+ "barge_in_detector/training/metrics.py": "6909e1cf3e4ac6b763c97ea9a8c26b5ee751a01fffc86c308604d5e437bbfad6",
26
+ "barge_in_detector/training/split.py": "44e05118111617a5649d94d09e82228864f2f2f5405a15dde974db78d9829a29",
27
+ "barge_in_detector/training/train.py": "a643f92b5c99d7c318e460dc2fdb5ae0bbb37003a8dae28fc3be60f8d37d1bd2",
28
+ "bin/create-labnotes": "5306e747935b105f878fa5f144c83595debb3f32f87c7d097720975bb1729884",
29
+ "data/export/utterances.csv": "556a2820e4a24ac13bab44321fb32b464c5fb225750d41a5a10ebcde7cdbab8a",
30
+ "data/splits/test.csv": "8b630782362b05249b3be94b0213850258422be5ddeb62fcf703817b0bf9e11a",
31
+ "data/splits/train.csv": "aab224bd0526eee915295e6f5c52076e29bb751cd0d664eefb4321843ad22533",
32
+ "data/splits/validation.csv": "c6b4c526fb4b1df96754119e409cedc8c7e85c77659e1fc1c3829da70084edc1",
33
+ "docs/data-format.md": "5d14d3aac1e432157c03bdf10cabf87e7366164629c70fb5904d632f472254ff",
34
+ "docs/factor-catalogue.md": "651e27021557eb7ce08c3ae5af0e341dee1aba665c1b6f72ad19625768167ec5",
35
+ "docs/generation.md": "1aae32dc0cb9c35904f8b2c5721ce7910494a27384a97b5f40fe0373c2d70b33",
36
+ "docs/training.md": "a2a96447aa065e4cb9f1e8a4b9ec97041d41ee32736cb68115706638b562d68f",
37
+ "docs/transcript-format.md": "9927370e6fb906d1dd1fabd98888d7e95d8ba2f373e7a1248e430f260c57aa32",
38
+ "docs/utterance-first-decision.md": "6c9ab89b7d579d83769e09eb3b77bf1cfa70434fe0cc45e24aaa959385a064e5",
39
+ "generation/config.yaml": "f4e707d4efc610b882289f29f170cd763c876c4e366f6853d1225cbb5965c88b",
40
+ "generation/factors/catalogue.yaml": "1135f3a3bf6000eb1f9d41b1957a348c760e78f96fbe61a8829aa818be0de5b5",
41
+ "generation/factors/constraints.yaml": "422318e1010483e4c5f7eb0de99170eb906b8d21927183a5d6581f13c21ac59c",
42
+ "generation/factors/domains.yaml": "19605dbdd0bb4044c481d74ce5c798534d61dd6c7c46d4edcb2f800c4aa53fc5",
43
+ "generation/factors/intents.yaml": "a3d343addabde711a2395e1ed598f432dd00e547afa962255ae246e17874ae63",
44
+ "generation/factors/languages.yaml": "c5f0a5b753c588782c536108184b12615a54c9b8f4b4cbf0073c0d8aec1a376e",
45
+ "generation/factors/utterances.yaml": "a045546c59d59588a261b03e1f1dc9dc147146f93d3381a813c0920825c435c4",
46
+ "generation/plans/english.yaml": "f1e7e17e2249bb736cd5cb75b1af0f0babbe3b2320e12b2b812346ff5c12d725",
47
+ "generation/plans/utterance-all.yaml": "20349fb83cd17d067736e4d26d746be7a77fa7cd7a8d41569fc44335cefb0a5d",
48
+ "generation/plans/utterance-sample.yaml": "7ab683e2c38af4f0105c957971efa2caaeca17c7bc2e11397e362591d0a31276",
49
+ "generation/prompts/generator_utterances_v5.md": "0a2e3eeeb7b7c7ff1ad2e2de8973af5469fcb8c369d9092d251c7590477b1266",
50
+ "labnotes/20261114-0907-project-brief-setup.md": "ab069c92129480d35758375cf5c57cbce96a643f007ed1d7d5941a6718bf0c51",
51
+ "labnotes/20261114-1023-factor-catalogue-draft.md": "6e3c2261c5bdb325548f59763ababe178d1f442f6440497c47d53c2773b513ee",
52
+ "labnotes/20261114-1541-synthetic-data-generator.md": "1b5f4bb84e3c224164d4e295b7639aad3a27d82d01948220b0c177fd1e93ea2b",
53
+ "labnotes/20261115-0100-utterance-first-generation.md": "5e81c46e2799a970b783e660b06ed16fdf4ea7bc43859887a4ae2d42fe5f6d0f",
54
+ "labnotes/20261115-1211-vastai-model-training.md": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
55
+ "labnotes/issues/.gitkeep": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
56
+ "labnotes/milestones/index.md": "0f3b3889fbbb7ea869c3c4e264d1da7cbd4535b28890113f35edcb8cf39b7463",
57
+ "labnotes/milestones/multilingual-text-barge-in-classifier.md": "c292dbefd18e5056cf219f8bd5c66403bcf291eab0cb9589840beb198ce5a74d",
58
+ "labnotes/milestones/utterance-first-backchannel-classifier.md": "e3bf3bd0ce94ac80d5b3ac4d697535cdb95fda30d7d4d1cad282b2d7e40c0d16",
59
+ "pyproject.toml": "4aef3a80e879d58c4c559b1f276e4b51c57006583cce3fd7a302e91336d9bbef",
60
+ "tests/fake_llm.py": "2c604f09ff23619225e0c62a57466f300191fd8391777ac7cf05e5556dee4f33",
61
+ "tests/fixtures/cleaner_golden.yaml": "3e013d3775a619a08432fe6c0ced5558934e4dbcf98a4afa651a0bd4bfc1c1db",
62
+ "tests/test_catalogue.py": "9fbfc7bc848e5db38d34ccb08f6bf10bd0908b21cbcc73d70d337c1e0dbae2ca",
63
+ "tests/test_cleaner.py": "048942dafd6e91772d7b17515606cb6c7f44c2225b876542d8598d19116a3d47",
64
+ "tests/test_gemini_live.py": "64f4ba3a80503ad7e0a43bfbaf7e43fbecbbc5821a41b189c08da5eff390e884",
65
+ "tests/test_ids.py": "3802dc4bd8b75f8764c1d4eafbc882b614096d5d871ec4f5d001990bb3f87d89",
66
+ "tests/test_llm.py": "eb8255bc88a0d9b8dde268ae67d700f5e2406c87b82cbda27dbba920ba9ad19a",
67
+ "tests/test_metrics.py": "072656cb74e8750f1a9a52f5d6dadb2ac3250e5f1cd538d6e64f3bd5a46d9567",
68
+ "tests/test_split.py": "d4556a43db6f706e6bac282bb5749792ddc66f12ba439d0b7fa1ad6cf666ff63",
69
+ "tests/test_store.py": "f23ab069512838929f7f0f91f8439ceaac4ba38484a6b235750fd7adf672614d",
70
+ "tests/test_utterance_criteria.py": "bb076041b038a6b707a6df61ee708408bec89922c011fef8dce0824991691a53",
71
+ "tests/test_utterances.py": "3a48f590d83cddc4db2f57099f5f90bff3b965195f0aeb1adcd663a27bc1a387",
72
+ "uv.lock": "3992e539e9e18aeb3182978ef9ef0b6ae8c0e15d649416e2de11b9ba98317ba8"
73
+ }
reports/training-metrics.json ADDED
@@ -0,0 +1,519 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "args": {
3
+ "out": "models/e5-small-run1",
4
+ "model": "intfloat/multilingual-e5-small",
5
+ "splits": "data/splits",
6
+ "epochs": "5",
7
+ "batch_size": "32",
8
+ "lr": "3e-05",
9
+ "warmup": "0.1",
10
+ "max_length": "64",
11
+ "seed": "20261115"
12
+ },
13
+ "best": {
14
+ "epoch": 5,
15
+ "train_loss": 0.02041708185817256,
16
+ "seconds": 1.7,
17
+ "validation": {
18
+ "count": 333,
19
+ "accuracy": 0.993993993993994,
20
+ "macro_f1": 0.9932432432432432,
21
+ "false_interruption_rate": 0.009009009009009009,
22
+ "missed_interruption_rate": 0.0045045045045045045,
23
+ "per_class": {
24
+ "backchannel": {
25
+ "precision": 0.990990990990991,
26
+ "recall": 0.990990990990991,
27
+ "f1": 0.990990990990991,
28
+ "count": 111
29
+ },
30
+ "barge_in": {
31
+ "precision": 0.9954954954954955,
32
+ "recall": 0.9954954954954955,
33
+ "f1": 0.9954954954954955,
34
+ "count": 222
35
+ }
36
+ },
37
+ "per_language": {
38
+ "de": {
39
+ "accuracy": 1.0,
40
+ "count": 18
41
+ },
42
+ "en": {
43
+ "accuracy": 1.0,
44
+ "count": 81
45
+ },
46
+ "es": {
47
+ "accuracy": 1.0,
48
+ "count": 36
49
+ },
50
+ "fil": {
51
+ "accuracy": 1.0,
52
+ "count": 18
53
+ },
54
+ "fr": {
55
+ "accuracy": 1.0,
56
+ "count": 18
57
+ },
58
+ "hi": {
59
+ "accuracy": 1.0,
60
+ "count": 18
61
+ },
62
+ "id": {
63
+ "accuracy": 1.0,
64
+ "count": 18
65
+ },
66
+ "it": {
67
+ "accuracy": 1.0,
68
+ "count": 18
69
+ },
70
+ "ja": {
71
+ "accuracy": 1.0,
72
+ "count": 18
73
+ },
74
+ "kn": {
75
+ "accuracy": 1.0,
76
+ "count": 18
77
+ },
78
+ "ml": {
79
+ "accuracy": 1.0,
80
+ "count": 18
81
+ },
82
+ "ta": {
83
+ "accuracy": 0.9444444444444444,
84
+ "count": 18
85
+ },
86
+ "te": {
87
+ "accuracy": 1.0,
88
+ "count": 18
89
+ },
90
+ "zh": {
91
+ "accuracy": 0.9444444444444444,
92
+ "count": 18
93
+ }
94
+ }
95
+ }
96
+ },
97
+ "history": [
98
+ {
99
+ "epoch": 1,
100
+ "train_loss": 0.48989867342240884,
101
+ "seconds": 2.3,
102
+ "validation": {
103
+ "count": 333,
104
+ "accuracy": 0.984984984984985,
105
+ "macro_f1": 0.98306980527734,
106
+ "false_interruption_rate": 0.02702702702702703,
107
+ "missed_interruption_rate": 0.009009009009009009,
108
+ "per_class": {
109
+ "backchannel": {
110
+ "precision": 0.9818181818181818,
111
+ "recall": 0.972972972972973,
112
+ "f1": 0.9773755656108598,
113
+ "count": 111
114
+ },
115
+ "barge_in": {
116
+ "precision": 0.9865470852017937,
117
+ "recall": 0.990990990990991,
118
+ "f1": 0.9887640449438202,
119
+ "count": 222
120
+ }
121
+ },
122
+ "per_language": {
123
+ "de": {
124
+ "accuracy": 0.9444444444444444,
125
+ "count": 18
126
+ },
127
+ "en": {
128
+ "accuracy": 1.0,
129
+ "count": 81
130
+ },
131
+ "es": {
132
+ "accuracy": 1.0,
133
+ "count": 36
134
+ },
135
+ "fil": {
136
+ "accuracy": 1.0,
137
+ "count": 18
138
+ },
139
+ "fr": {
140
+ "accuracy": 1.0,
141
+ "count": 18
142
+ },
143
+ "hi": {
144
+ "accuracy": 1.0,
145
+ "count": 18
146
+ },
147
+ "id": {
148
+ "accuracy": 1.0,
149
+ "count": 18
150
+ },
151
+ "it": {
152
+ "accuracy": 1.0,
153
+ "count": 18
154
+ },
155
+ "ja": {
156
+ "accuracy": 1.0,
157
+ "count": 18
158
+ },
159
+ "kn": {
160
+ "accuracy": 0.9444444444444444,
161
+ "count": 18
162
+ },
163
+ "ml": {
164
+ "accuracy": 0.9444444444444444,
165
+ "count": 18
166
+ },
167
+ "ta": {
168
+ "accuracy": 0.9444444444444444,
169
+ "count": 18
170
+ },
171
+ "te": {
172
+ "accuracy": 1.0,
173
+ "count": 18
174
+ },
175
+ "zh": {
176
+ "accuracy": 0.9444444444444444,
177
+ "count": 18
178
+ }
179
+ }
180
+ }
181
+ },
182
+ {
183
+ "epoch": 2,
184
+ "train_loss": 0.08939109126275235,
185
+ "seconds": 1.7,
186
+ "validation": {
187
+ "count": 333,
188
+ "accuracy": 0.990990990990991,
189
+ "macro_f1": 0.9898875380862241,
190
+ "false_interruption_rate": 0.009009009009009009,
191
+ "missed_interruption_rate": 0.009009009009009009,
192
+ "per_class": {
193
+ "backchannel": {
194
+ "precision": 0.9821428571428571,
195
+ "recall": 0.990990990990991,
196
+ "f1": 0.9865470852017937,
197
+ "count": 111
198
+ },
199
+ "barge_in": {
200
+ "precision": 0.995475113122172,
201
+ "recall": 0.990990990990991,
202
+ "f1": 0.9932279909706545,
203
+ "count": 222
204
+ }
205
+ },
206
+ "per_language": {
207
+ "de": {
208
+ "accuracy": 1.0,
209
+ "count": 18
210
+ },
211
+ "en": {
212
+ "accuracy": 1.0,
213
+ "count": 81
214
+ },
215
+ "es": {
216
+ "accuracy": 1.0,
217
+ "count": 36
218
+ },
219
+ "fil": {
220
+ "accuracy": 1.0,
221
+ "count": 18
222
+ },
223
+ "fr": {
224
+ "accuracy": 1.0,
225
+ "count": 18
226
+ },
227
+ "hi": {
228
+ "accuracy": 1.0,
229
+ "count": 18
230
+ },
231
+ "id": {
232
+ "accuracy": 1.0,
233
+ "count": 18
234
+ },
235
+ "it": {
236
+ "accuracy": 1.0,
237
+ "count": 18
238
+ },
239
+ "ja": {
240
+ "accuracy": 1.0,
241
+ "count": 18
242
+ },
243
+ "kn": {
244
+ "accuracy": 1.0,
245
+ "count": 18
246
+ },
247
+ "ml": {
248
+ "accuracy": 0.9444444444444444,
249
+ "count": 18
250
+ },
251
+ "ta": {
252
+ "accuracy": 0.9444444444444444,
253
+ "count": 18
254
+ },
255
+ "te": {
256
+ "accuracy": 1.0,
257
+ "count": 18
258
+ },
259
+ "zh": {
260
+ "accuracy": 0.9444444444444444,
261
+ "count": 18
262
+ }
263
+ }
264
+ }
265
+ },
266
+ {
267
+ "epoch": 3,
268
+ "train_loss": 0.03671385896476832,
269
+ "seconds": 1.7,
270
+ "validation": {
271
+ "count": 333,
272
+ "accuracy": 0.987987987987988,
273
+ "macro_f1": 0.9865465416936006,
274
+ "false_interruption_rate": 0.009009009009009009,
275
+ "missed_interruption_rate": 0.013513513513513514,
276
+ "per_class": {
277
+ "backchannel": {
278
+ "precision": 0.9734513274336283,
279
+ "recall": 0.990990990990991,
280
+ "f1": 0.9821428571428571,
281
+ "count": 111
282
+ },
283
+ "barge_in": {
284
+ "precision": 0.9954545454545455,
285
+ "recall": 0.9864864864864865,
286
+ "f1": 0.9909502262443439,
287
+ "count": 222
288
+ }
289
+ },
290
+ "per_language": {
291
+ "de": {
292
+ "accuracy": 1.0,
293
+ "count": 18
294
+ },
295
+ "en": {
296
+ "accuracy": 1.0,
297
+ "count": 81
298
+ },
299
+ "es": {
300
+ "accuracy": 1.0,
301
+ "count": 36
302
+ },
303
+ "fil": {
304
+ "accuracy": 1.0,
305
+ "count": 18
306
+ },
307
+ "fr": {
308
+ "accuracy": 1.0,
309
+ "count": 18
310
+ },
311
+ "hi": {
312
+ "accuracy": 1.0,
313
+ "count": 18
314
+ },
315
+ "id": {
316
+ "accuracy": 1.0,
317
+ "count": 18
318
+ },
319
+ "it": {
320
+ "accuracy": 1.0,
321
+ "count": 18
322
+ },
323
+ "ja": {
324
+ "accuracy": 1.0,
325
+ "count": 18
326
+ },
327
+ "kn": {
328
+ "accuracy": 1.0,
329
+ "count": 18
330
+ },
331
+ "ml": {
332
+ "accuracy": 1.0,
333
+ "count": 18
334
+ },
335
+ "ta": {
336
+ "accuracy": 0.9444444444444444,
337
+ "count": 18
338
+ },
339
+ "te": {
340
+ "accuracy": 1.0,
341
+ "count": 18
342
+ },
343
+ "zh": {
344
+ "accuracy": 0.8333333333333334,
345
+ "count": 18
346
+ }
347
+ }
348
+ }
349
+ },
350
+ {
351
+ "epoch": 4,
352
+ "train_loss": 0.025003968517888674,
353
+ "seconds": 1.7,
354
+ "validation": {
355
+ "count": 333,
356
+ "accuracy": 0.987987987987988,
357
+ "macro_f1": 0.9865465416936006,
358
+ "false_interruption_rate": 0.009009009009009009,
359
+ "missed_interruption_rate": 0.013513513513513514,
360
+ "per_class": {
361
+ "backchannel": {
362
+ "precision": 0.9734513274336283,
363
+ "recall": 0.990990990990991,
364
+ "f1": 0.9821428571428571,
365
+ "count": 111
366
+ },
367
+ "barge_in": {
368
+ "precision": 0.9954545454545455,
369
+ "recall": 0.9864864864864865,
370
+ "f1": 0.9909502262443439,
371
+ "count": 222
372
+ }
373
+ },
374
+ "per_language": {
375
+ "de": {
376
+ "accuracy": 1.0,
377
+ "count": 18
378
+ },
379
+ "en": {
380
+ "accuracy": 1.0,
381
+ "count": 81
382
+ },
383
+ "es": {
384
+ "accuracy": 1.0,
385
+ "count": 36
386
+ },
387
+ "fil": {
388
+ "accuracy": 1.0,
389
+ "count": 18
390
+ },
391
+ "fr": {
392
+ "accuracy": 1.0,
393
+ "count": 18
394
+ },
395
+ "hi": {
396
+ "accuracy": 1.0,
397
+ "count": 18
398
+ },
399
+ "id": {
400
+ "accuracy": 1.0,
401
+ "count": 18
402
+ },
403
+ "it": {
404
+ "accuracy": 1.0,
405
+ "count": 18
406
+ },
407
+ "ja": {
408
+ "accuracy": 1.0,
409
+ "count": 18
410
+ },
411
+ "kn": {
412
+ "accuracy": 1.0,
413
+ "count": 18
414
+ },
415
+ "ml": {
416
+ "accuracy": 1.0,
417
+ "count": 18
418
+ },
419
+ "ta": {
420
+ "accuracy": 0.9444444444444444,
421
+ "count": 18
422
+ },
423
+ "te": {
424
+ "accuracy": 1.0,
425
+ "count": 18
426
+ },
427
+ "zh": {
428
+ "accuracy": 0.8333333333333334,
429
+ "count": 18
430
+ }
431
+ }
432
+ }
433
+ },
434
+ {
435
+ "epoch": 5,
436
+ "train_loss": 0.02041708185817256,
437
+ "seconds": 1.7,
438
+ "validation": {
439
+ "count": 333,
440
+ "accuracy": 0.993993993993994,
441
+ "macro_f1": 0.9932432432432432,
442
+ "false_interruption_rate": 0.009009009009009009,
443
+ "missed_interruption_rate": 0.0045045045045045045,
444
+ "per_class": {
445
+ "backchannel": {
446
+ "precision": 0.990990990990991,
447
+ "recall": 0.990990990990991,
448
+ "f1": 0.990990990990991,
449
+ "count": 111
450
+ },
451
+ "barge_in": {
452
+ "precision": 0.9954954954954955,
453
+ "recall": 0.9954954954954955,
454
+ "f1": 0.9954954954954955,
455
+ "count": 222
456
+ }
457
+ },
458
+ "per_language": {
459
+ "de": {
460
+ "accuracy": 1.0,
461
+ "count": 18
462
+ },
463
+ "en": {
464
+ "accuracy": 1.0,
465
+ "count": 81
466
+ },
467
+ "es": {
468
+ "accuracy": 1.0,
469
+ "count": 36
470
+ },
471
+ "fil": {
472
+ "accuracy": 1.0,
473
+ "count": 18
474
+ },
475
+ "fr": {
476
+ "accuracy": 1.0,
477
+ "count": 18
478
+ },
479
+ "hi": {
480
+ "accuracy": 1.0,
481
+ "count": 18
482
+ },
483
+ "id": {
484
+ "accuracy": 1.0,
485
+ "count": 18
486
+ },
487
+ "it": {
488
+ "accuracy": 1.0,
489
+ "count": 18
490
+ },
491
+ "ja": {
492
+ "accuracy": 1.0,
493
+ "count": 18
494
+ },
495
+ "kn": {
496
+ "accuracy": 1.0,
497
+ "count": 18
498
+ },
499
+ "ml": {
500
+ "accuracy": 1.0,
501
+ "count": 18
502
+ },
503
+ "ta": {
504
+ "accuracy": 0.9444444444444444,
505
+ "count": 18
506
+ },
507
+ "te": {
508
+ "accuracy": 1.0,
509
+ "count": 18
510
+ },
511
+ "zh": {
512
+ "accuracy": 0.9444444444444444,
513
+ "count": 18
514
+ }
515
+ }
516
+ }
517
+ }
518
+ ]
519
+ }
tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:255f5e32cb320b4dc2b31e0f203832fecafe87ca4cc87c3c41eb8ccae73f441a
3
+ size 17082987
tokenizer_config.json ADDED
@@ -0,0 +1,17 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": true,
3
+ "backend": "tokenizers",
4
+ "bos_token": "<s>",
5
+ "clean_up_tokenization_spaces": true,
6
+ "cls_token": "<s>",
7
+ "eos_token": "</s>",
8
+ "is_local": false,
9
+ "local_files_only": false,
10
+ "mask_token": "<mask>",
11
+ "model_max_length": 512,
12
+ "pad_token": "<pad>",
13
+ "sep_token": "</s>",
14
+ "sp_model_kwargs": {},
15
+ "tokenizer_class": "XLMRobertaTokenizer",
16
+ "unk_token": "<unk>"
17
+ }