Hugging Face's logo

moondream
/
parakeet-redux

Automatic Speech Recognition
Safetensors
parakeet_tdt
ternary
parakeet
tdt
speech-recognition
cpu
apple-silicon
Eval Results
8-bit precision
Model card Files
xet
 Community
2
vikhyatk commited on
Commit
2bf1286
·
0 Parent(s):

parakeet-redux: weights, config, tokenizer and model card

Browse files
Files changed (7) hide show
  1. .eval_results/open_asr_leaderboard.yaml +96 -0
  2. .gitattributes +35 -0
  3. README.md +212 -0
  4. config.json +312 -0
  5. model.safetensors +3 -0
  6. ternary.json +2417 -0
  7. tokenizer.json +0 -0
.eval_results/open_asr_leaderboard.yaml ADDED
@@ -0,0 +1,96 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Self-reported. Word error rate in percent, scored with the Open ASR Leaderboard's own pipeline (its normalizers as of
2
+ # September 2026) on the leaderboard's test sets, through Photon (https://moondream.ai/photon). Details in the model
3
+ # card. The leaderboard's TED-LIUM split is no longer distributed; tedlium_wer scores the same eleven TED-LIUM 3 test
4
+ # talks in full (the card's long-form set), and mean_wer averages all eight sets. The card's 'Open ASR' figure (6.55)
5
+ # is the mean of the other seven.
6
+
7
+ - dataset:
8
+ id: hf-audio/open-asr-leaderboard
9
+ task_id: mean_wer
10
+ value: 6.05
11
+ date: '2026-09-21'
12
+ source:
13
+ url: https://huggingface.co/moondream/parakeet-redux
14
+ name: moondream
15
+ user: moondream
16
+
17
+ - dataset:
18
+ id: hf-audio/open-asr-leaderboard
19
+ task_id: ami_wer
20
+ value: 10.80
21
+ date: '2026-09-21'
22
+ source:
23
+ url: https://huggingface.co/moondream/parakeet-redux
24
+ name: moondream
25
+ user: moondream
26
+
27
+ - dataset:
28
+ id: hf-audio/open-asr-leaderboard
29
+ task_id: earnings22_wer
30
+ value: 9.95
31
+ date: '2026-09-21'
32
+ source:
33
+ url: https://huggingface.co/moondream/parakeet-redux
34
+ name: moondream
35
+ user: moondream
36
+
37
+ - dataset:
38
+ id: hf-audio/open-asr-leaderboard
39
+ task_id: gigaspeech_wer
40
+ value: 8.73
41
+ date: '2026-09-21'
42
+ source:
43
+ url: https://huggingface.co/moondream/parakeet-redux
44
+ name: moondream
45
+ user: moondream
46
+
47
+ - dataset:
48
+ id: hf-audio/open-asr-leaderboard
49
+ task_id: librispeech_clean_wer
50
+ value: 1.96
51
+ date: '2026-09-21'
52
+ source:
53
+ url: https://huggingface.co/moondream/parakeet-redux
54
+ name: moondream
55
+ user: moondream
56
+
57
+ - dataset:
58
+ id: hf-audio/open-asr-leaderboard
59
+ task_id: librispeech_other_wer
60
+ value: 4.34
61
+ date: '2026-09-21'
62
+ source:
63
+ url: https://huggingface.co/moondream/parakeet-redux
64
+ name: moondream
65
+ user: moondream
66
+
67
+ - dataset:
68
+ id: hf-audio/open-asr-leaderboard
69
+ task_id: spgispeech_wer
70
+ value: 4.01
71
+ date: '2026-09-21'
72
+ source:
73
+ url: https://huggingface.co/moondream/parakeet-redux
74
+ name: moondream
75
+ user: moondream
76
+
77
+ - dataset:
78
+ id: hf-audio/open-asr-leaderboard
79
+ task_id: tedlium_wer
80
+ # the eleven TED-LIUM 3 test talks scored whole; see the note above
81
+ value: 2.51
82
+ date: '2026-09-21'
83
+ source:
84
+ url: https://huggingface.co/moondream/parakeet-redux
85
+ name: moondream
86
+ user: moondream
87
+
88
+ - dataset:
89
+ id: hf-audio/open-asr-leaderboard
90
+ task_id: voxpopuli_wer
91
+ value: 6.07
92
+ date: '2026-09-21'
93
+ source:
94
+ url: https://huggingface.co/moondream/parakeet-redux
95
+ name: moondream
96
+ user: moondream
.gitattributes ADDED
@@ -0,0 +1,35 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
README.md ADDED
@@ -0,0 +1,212 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-4.0
3
+ language: [en, de, fr, es, it, pt, ru, uk, hr, sl, lv, lt, et, fi, sv, da, nl, pl, cs, sk, hu, ro, bg, el, mt]
4
+ pipeline_tag: automatic-speech-recognition
5
+ tags: [ternary, parakeet, tdt, speech-recognition, cpu, apple-silicon]
6
+ ---
7
+
8
+ # Moondream Parakeet Redux
9
+
10
+ A 1.58-bit version of [parakeet-tdt-0.6b-v3](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3). Same architecture,
11
+ same tokenizer, but every encoder weight is -1, 0 or +1. It fits in 178 MB, runs at 113× real
12
+ time on eight x86 CPU cores, 2.5× the fastest other Parakeet runtime we measured, and stays within 0.3 WER of the
13
+ original on English while beating it on the 25-language FLEURS set and on long-form audio.
14
+
15
+ | | parakeet-tdt-0.6b-v3 | parakeet-redux |
16
+ |---|---|---|
17
+ | Open ASR Leaderboard, 7 English sets (WER %) | **6.26** | 6.55 |
18
+ | FLEURS, 25 languages (WER %) | 11.62 | **10.56** |
19
+ | Business speech, AA-WER style (WER %) | **6.15** | 6.96 |
20
+ | Background noise, 9 MUSAN conditions (WER %) | **6.72** | 9.04 |
21
+ | TED-LIUM long-form (WER %) | 2.71 | **2.51** |
22
+ | Weights | 1.2 GB | **178 MB** |
23
+
24
+ Its sibling [Parakeet Ultra](https://huggingface.co/moondream/parakeet-ultra) is the full-precision version of the
25
+ same architecture, trained further, for GPUs: better than the original on every benchmark. Read the
26
+ [release post](https://moondream.ai/blog/introducing-parakeet-redux-and-ultra) for the story behind both models.
27
+
28
+ ## Usage
29
+
30
+ Run it with [Photon](https://moondream.ai/photon), which reads the packed weights directly: AVX-512 VNNI on x86,
31
+ NEON on ARM, Metal on Apple GPUs. Every speed number on this page is Photon.
32
+
33
+ ```python
34
+ # pip install moondream (2.4.0 or later)
35
+ import moondream as md
36
+
37
+ # device: "cpu", "mps" or "cuda"; leave it out to take CUDA,
38
+ # then Apple silicon, then the CPU
39
+ with md.photon("moondream/parakeet-redux", device="cpu") as speech:
40
+ result = speech.transcribe(audio="speech.wav")
41
+ print(result["text"])
42
+ # From the undersurface of the clouds there are continual
43
+ # emissions of lurid light. Electric matter is in continual
44
+ # evolution from their component molecules. The gaseous ...
45
+
46
+ # timestamps="segment": one entry per sentence, with its start
47
+ # and end in seconds
48
+ result = speech.transcribe(audio="speech.wav", timestamps="segment")
49
+ for segment in result["segments"]:
50
+ print(segment["start"], segment["end"], segment["text"])
51
+ # 0.32 5.12 From the undersurface of the clouds there are ...
52
+ # 5.12 9.92 Electric matter is in continual evolution from ...
53
+ # 9.92 21.04 The gaseous elements of the air need to be ...
54
+
55
+ # timestamps="word": the same sentences, each with the start
56
+ # and end of every word
57
+ result = speech.transcribe(audio="speech.wav", timestamps="word")
58
+ for word in result["segments"][0]["words"][:3]:
59
+ print(word["start"], word["end"], word["word"])
60
+ # 0.32 0.56 From
61
+ # 0.56 0.72 the
62
+ # 0.72 1.44 undersurface
63
+ ```
64
+
65
+ ## Performance
66
+
67
+ Real-time factor: seconds of audio transcribed per second of wall clock, one utterance at a time, higher is
68
+ faster. The other rows are the fastest other ways we found to run Parakeet on the same machine and the same audio.
69
+
70
+ ### How this was measured
71
+
72
+ One utterance at a time. Incumbents: parakeet.cpp (ggml), sherpa-onnx and onnx-asr (ONNX Runtime) and, on the
73
+ Mac, parakeet-mlx, each at its own defaults. x86: LibriSpeech test-clean (2,620 utterances), every runtime on the
74
+ same 8 cores. Apple silicon: a 50-utterance slice of LibriSpeech dev-clean, with cool-downs between runs. WER is
75
+ scored the same way as the benchmarks below.
76
+
77
+ ### x86 CPU
78
+
79
+ AMD EPYC 9575F (Zen 5, up to 5.0 GHz, AVX-512), 8 physical cores of one chiplet, DDR5-6000, Ubuntu 22.04.
80
+
81
+ | runtime | weights | real time | WER |
82
+ |---|---|---|---|
83
+ | **Photon, this model** | ternary, 178 MB | **113×** | 1.94 |
84
+ | parakeet.cpp (ggml) | q8_0, 0.94 GB | 45× | 1.51 |
85
+ | sherpa-onnx (ONNX Runtime) | int8, 0.67 GB | 42× | 1.97 |
86
+ | onnx-asr (ONNX Runtime) | int8, 0.67 GB | 28× | 1.93 |
87
+
88
+ ### Apple silicon
89
+
90
+ MacBook Air with the Apple M2 (4 performance and 4 efficiency CPU cores, 10-core GPU), 16 GB unified memory, macOS 15.
91
+
92
+ | runtime | weights | CPU | GPU |
93
+ |---|---|---|---|
94
+ | **Photon, this model** | ternary, 178 MB | **38×** | **43×** |
95
+ | parakeet.cpp (ggml) | q8_0, 0.94 GB | 12× | 38× (Metal) |
96
+ | parakeet.cpp (ggml) | f16, 1.44 GB | 9× | 39× (Metal) |
97
+ | parakeet-mlx | fp32, 2.51 GB | — | 37× |
98
+ | onnx-asr (ONNX Runtime) | int8, 0.67 GB | 33× | — |
99
+ | sherpa-onnx (ONNX Runtime) | int8, 0.67 GB | 28× | — |
100
+
101
+ ## Benchmarks
102
+
103
+ Word error rate in percent, lower is better. parakeet-redux is better than the original on FLEURS and on
104
+ long-form audio, close to it on English, and behind it in noise. Both models are scored on the same files with the
105
+ Open ASR Leaderboard's own pipeline as of September 2026 (its normalizers and compound-merging alignment, with the
106
+ FLEURS references prepared as the leaderboard's text column is). parakeet-redux runs in Photon on an NVIDIA GPU, the
107
+ original in NeMo in bf16.
108
+
109
+ ### Open ASR Leaderboard
110
+
111
+ The seven English test sets of the Hugging Face Open ASR Leaderboard: audiobooks (LibriSpeech), meetings (AMI),
112
+ earnings calls (Earnings-22), podcasts and YouTube (GigaSpeech), financial calls (SPGISpeech) and parliament
113
+ (VoxPopuli).
114
+
115
+ | set | parakeet-tdt-0.6b-v3 | parakeet-redux |
116
+ |---|---|---|
117
+ | LibriSpeech test-clean | 1.52 | 1.96 |
118
+ | LibriSpeech test-other | 3.13 | 4.34 |
119
+ | AMI | 10.86 | 10.80 |
120
+ | Earnings-22 | 10.75 | 9.95 |
121
+ | GigaSpeech | 8.05 | 8.73 |
122
+ | SPGISpeech | 3.63 | 4.01 |
123
+ | VoxPopuli | 5.88 | 6.07 |
124
+ | **average** | **6.26** | **6.55** |
125
+
126
+ ### FLEURS
127
+
128
+ The FLEURS test split: read Wikipedia sentences, a few hundred per language, for all 25 languages the model
129
+ supports.
130
+
131
+ | language | parakeet-tdt-0.6b-v3 | parakeet-redux |
132
+ |---|---|---|
133
+ | Bulgarian | 11.90 | 11.23 |
134
+ | Croatian | 10.93 | 9.26 |
135
+ | Czech | 10.85 | 10.25 |
136
+ | Danish | 16.78 | 15.94 |
137
+ | Dutch | 6.18 | 7.45 |
138
+ | English | 4.25 | 4.90 |
139
+ | Estonian | 13.23 | 9.15 |
140
+ | Finnish | 11.05 | 10.38 |
141
+ | French | 4.81 | 7.71 |
142
+ | German | 4.13 | 5.42 |
143
+ | Greek | 35.71 | 32.48 |
144
+ | Hungarian | 13.65 | 14.15 |
145
+ | Italian | 2.61 | 3.24 |
146
+ | Latvian | 21.38 | 12.80 |
147
+ | Lithuanian | 21.09 | 17.27 |
148
+ | Maltese | 19.13 | 13.65 |
149
+ | Polish | 6.70 | 8.59 |
150
+ | Portuguese | 4.65 | 4.99 |
151
+ | Romanian | 11.54 | 10.32 |
152
+ | Russian | 5.91 | 7.91 |
153
+ | Slovak | 9.46 | 7.26 |
154
+ | Slovene | 21.76 | 16.21 |
155
+ | Spanish | 3.12 | 3.71 |
156
+ | Swedish | 13.75 | 12.71 |
157
+ | Ukrainian | 5.94 | 7.06 |
158
+ | **average** | **11.62** | **10.56** |
159
+
160
+ ### Business speech (AA-WER style)
161
+
162
+ The AMI and VoxPopuli test sets with the Artificial Analysis cleaning applied, and Earnings-22 scored in 30-second
163
+ chunks joined per call, the way the AA-WER benchmark does it.
164
+
165
+ | set | parakeet-tdt-0.6b-v3 | parakeet-redux |
166
+ |---|---|---|
167
+ | AMI (cleaned) | 9.52 | 9.14 |
168
+ | VoxPopuli (cleaned) | 3.02 | 3.86 |
169
+ | Earnings-22, 30-second chunks | 5.90 | 7.89 |
170
+ | **average** | **6.15** | **6.96** |
171
+
172
+ ### Background noise
173
+
174
+ The clean sets with MUSAN noise (the half of the corpus not used for training) mixed in at a fixed signal-to-noise
175
+ ratio; 0 dB means the noise is as loud as the speech.
176
+
177
+ | set | parakeet-tdt-0.6b-v3 | parakeet-redux |
178
+ |---|---|---|
179
+ | LibriSpeech test-other, 10 dB | 4.12 | 5.66 |
180
+ | LibriSpeech test-other, 5 dB | 5.49 | 7.32 |
181
+ | LibriSpeech test-other, 0 dB | 9.06 | 10.95 |
182
+ | FLEURS German, 10 dB | 5.78 | 8.97 |
183
+ | FLEURS German, 5 dB | 8.07 | 12.36 |
184
+ | FLEURS German, 0 dB | 14.45 | 19.18 |
185
+ | FLEURS Spanish, 10 dB | 3.99 | 4.46 |
186
+ | FLEURS Spanish, 5 dB | 4.12 | 5.25 |
187
+ | FLEURS Spanish, 0 dB | 5.44 | 7.22 |
188
+ | **average** | **6.72** | **9.04** |
189
+
190
+ Noise is where the gap to the original is widest: the ternary encoder's acoustic margin is thinner, and at low SNR
191
+ it substitutes similar-sounding words more often. Dropped or invented content is not more frequent than the
192
+ original's.
193
+
194
+ ### Long-form
195
+
196
+ Eleven complete TED-LIUM 3 talks of 10-20 minutes each. parakeet-redux transcribes them through Photon, whose
197
+ segmenter cuts each talk at pauses found by the model's VAD head into segments of at most 30 seconds; the original
198
+ runs NeMo's own long-audio path.
199
+
200
+ | set | parakeet-tdt-0.6b-v3 | parakeet-redux |
201
+ |---|---|---|
202
+ | TED-LIUM 3, 11 full talks of 10-20 minutes | 2.71 | 2.51 |
203
+
204
+ ## Notes
205
+
206
+ - Based on [parakeet-tdt-0.6b-v3](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3) by NVIDIA. Languages,
207
+ tokenizer and output conventions (punctuation, casing, numerals) are the original's.
208
+ - Runs with [Photon](https://moondream.ai/photon).
209
+ - Long audio is segmented by the model itself: the weights carry a small voice-activity head on the encoder's
210
+ subsampler, and Photon uses it to cut recordings at pauses into segments of at most 30 seconds. No external VAD
211
+ model is needed.
212
+ - License is CC-BY-4.0, same as the original.
config.json ADDED
@@ -0,0 +1,312 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "blank_token_id": 8192,
3
+ "decoder_hidden_size": 640,
4
+ "dtype": "float32",
5
+ "durations": [
6
+ 0,
7
+ 1,
8
+ 2,
9
+ 3,
10
+ 4
11
+ ],
12
+ "encoder_config": {
13
+ "activation_dropout": 0.1,
14
+ "attention_bias": false,
15
+ "attention_dropout": 0.1,
16
+ "conv_kernel_size": 9,
17
+ "convolution_bias": false,
18
+ "dropout": 0.1,
19
+ "dropout_positions": 0.0,
20
+ "hidden_act": "silu",
21
+ "hidden_size": 1024,
22
+ "initializer_range": 0.02,
23
+ "intermediate_size": 4096,
24
+ "layerdrop": 0.1,
25
+ "max_position_embeddings": 5000,
26
+ "model_type": "parakeet_encoder",
27
+ "num_attention_heads": 8,
28
+ "num_hidden_layers": 24,
29
+ "num_key_value_heads": 8,
30
+ "num_mel_bins": 128,
31
+ "scale_input": false,
32
+ "subsampling_conv_channels": 256,
33
+ "subsampling_conv_kernel_size": 3,
34
+ "subsampling_conv_stride": 2,
35
+ "subsampling_factor": 8
36
+ },
37
+ "hidden_act": "relu",
38
+ "initializer_range": 0.02,
39
+ "is_encoder_decoder": true,
40
+ "max_symbols_per_step": 10,
41
+ "model_type": "parakeet_tdt",
42
+ "num_decoder_layers": 2,
43
+ "pad_token_id": 2,
44
+ "vocab_size": 8193,
45
+ "ternary_group_size": 128,
46
+ "ternary_modules": [
47
+ "encoder.layers.0.feed_forward1.linear1",
48
+ "encoder.layers.0.feed_forward1.linear2",
49
+ "encoder.layers.0.self_attn.q_proj",
50
+ "encoder.layers.0.self_attn.k_proj",
51
+ "encoder.layers.0.self_attn.v_proj",
52
+ "encoder.layers.0.self_attn.o_proj",
53
+ "encoder.layers.0.self_attn.relative_k_proj",
54
+ "encoder.layers.0.conv.pointwise_conv1",
55
+ "encoder.layers.0.conv.pointwise_conv2",
56
+ "encoder.layers.0.feed_forward2.linear1",
57
+ "encoder.layers.0.feed_forward2.linear2",
58
+ "encoder.layers.1.feed_forward1.linear1",
59
+ "encoder.layers.1.feed_forward1.linear2",
60
+ "encoder.layers.1.self_attn.q_proj",
61
+ "encoder.layers.1.self_attn.k_proj",
62
+ "encoder.layers.1.self_attn.v_proj",
63
+ "encoder.layers.1.self_attn.o_proj",
64
+ "encoder.layers.1.self_attn.relative_k_proj",
65
+ "encoder.layers.1.conv.pointwise_conv1",
66
+ "encoder.layers.1.conv.pointwise_conv2",
67
+ "encoder.layers.1.feed_forward2.linear1",
68
+ "encoder.layers.1.feed_forward2.linear2",
69
+ "encoder.layers.2.feed_forward1.linear1",
70
+ "encoder.layers.2.feed_forward1.linear2",
71
+ "encoder.layers.2.self_attn.q_proj",
72
+ "encoder.layers.2.self_attn.k_proj",
73
+ "encoder.layers.2.self_attn.v_proj",
74
+ "encoder.layers.2.self_attn.o_proj",
75
+ "encoder.layers.2.self_attn.relative_k_proj",
76
+ "encoder.layers.2.conv.pointwise_conv1",
77
+ "encoder.layers.2.conv.pointwise_conv2",
78
+ "encoder.layers.2.feed_forward2.linear1",
79
+ "encoder.layers.2.feed_forward2.linear2",
80
+ "encoder.layers.3.feed_forward1.linear1",
81
+ "encoder.layers.3.feed_forward1.linear2",
82
+ "encoder.layers.3.self_attn.q_proj",
83
+ "encoder.layers.3.self_attn.k_proj",
84
+ "encoder.layers.3.self_attn.v_proj",
85
+ "encoder.layers.3.self_attn.o_proj",
86
+ "encoder.layers.3.self_attn.relative_k_proj",
87
+ "encoder.layers.3.conv.pointwise_conv1",
88
+ "encoder.layers.3.conv.pointwise_conv2",
89
+ "encoder.layers.3.feed_forward2.linear1",
90
+ "encoder.layers.3.feed_forward2.linear2",
91
+ "encoder.layers.4.feed_forward1.linear1",
92
+ "encoder.layers.4.feed_forward1.linear2",
93
+ "encoder.layers.4.self_attn.q_proj",
94
+ "encoder.layers.4.self_attn.k_proj",
95
+ "encoder.layers.4.self_attn.v_proj",
96
+ "encoder.layers.4.self_attn.o_proj",
97
+ "encoder.layers.4.self_attn.relative_k_proj",
98
+ "encoder.layers.4.conv.pointwise_conv1",
99
+ "encoder.layers.4.conv.pointwise_conv2",
100
+ "encoder.layers.4.feed_forward2.linear1",
101
+ "encoder.layers.4.feed_forward2.linear2",
102
+ "encoder.layers.5.feed_forward1.linear1",
103
+ "encoder.layers.5.feed_forward1.linear2",
104
+ "encoder.layers.5.self_attn.q_proj",
105
+ "encoder.layers.5.self_attn.k_proj",
106
+ "encoder.layers.5.self_attn.v_proj",
107
+ "encoder.layers.5.self_attn.o_proj",
108
+ "encoder.layers.5.self_attn.relative_k_proj",
109
+ "encoder.layers.5.conv.pointwise_conv1",
110
+ "encoder.layers.5.conv.pointwise_conv2",
111
+ "encoder.layers.5.feed_forward2.linear1",
112
+ "encoder.layers.5.feed_forward2.linear2",
113
+ "encoder.layers.6.feed_forward1.linear1",
114
+ "encoder.layers.6.feed_forward1.linear2",
115
+ "encoder.layers.6.self_attn.q_proj",
116
+ "encoder.layers.6.self_attn.k_proj",
117
+ "encoder.layers.6.self_attn.v_proj",
118
+ "encoder.layers.6.self_attn.o_proj",
119
+ "encoder.layers.6.self_attn.relative_k_proj",
120
+ "encoder.layers.6.conv.pointwise_conv1",
121
+ "encoder.layers.6.conv.pointwise_conv2",
122
+ "encoder.layers.6.feed_forward2.linear1",
123
+ "encoder.layers.6.feed_forward2.linear2",
124
+ "encoder.layers.7.feed_forward1.linear1",
125
+ "encoder.layers.7.feed_forward1.linear2",
126
+ "encoder.layers.7.self_attn.q_proj",
127
+ "encoder.layers.7.self_attn.k_proj",
128
+ "encoder.layers.7.self_attn.v_proj",
129
+ "encoder.layers.7.self_attn.o_proj",
130
+ "encoder.layers.7.self_attn.relative_k_proj",
131
+ "encoder.layers.7.conv.pointwise_conv1",
132
+ "encoder.layers.7.conv.pointwise_conv2",
133
+ "encoder.layers.7.feed_forward2.linear1",
134
+ "encoder.layers.7.feed_forward2.linear2",
135
+ "encoder.layers.8.feed_forward1.linear1",
136
+ "encoder.layers.8.feed_forward1.linear2",
137
+ "encoder.layers.8.self_attn.q_proj",
138
+ "encoder.layers.8.self_attn.k_proj",
139
+ "encoder.layers.8.self_attn.v_proj",
140
+ "encoder.layers.8.self_attn.o_proj",
141
+ "encoder.layers.8.self_attn.relative_k_proj",
142
+ "encoder.layers.8.conv.pointwise_conv1",
143
+ "encoder.layers.8.conv.pointwise_conv2",
144
+ "encoder.layers.8.feed_forward2.linear1",
145
+ "encoder.layers.8.feed_forward2.linear2",
146
+ "encoder.layers.9.feed_forward1.linear1",
147
+ "encoder.layers.9.feed_forward1.linear2",
148
+ "encoder.layers.9.self_attn.q_proj",
149
+ "encoder.layers.9.self_attn.k_proj",
150
+ "encoder.layers.9.self_attn.v_proj",
151
+ "encoder.layers.9.self_attn.o_proj",
152
+ "encoder.layers.9.self_attn.relative_k_proj",
153
+ "encoder.layers.9.conv.pointwise_conv1",
154
+ "encoder.layers.9.conv.pointwise_conv2",
155
+ "encoder.layers.9.feed_forward2.linear1",
156
+ "encoder.layers.9.feed_forward2.linear2",
157
+ "encoder.layers.10.feed_forward1.linear1",
158
+ "encoder.layers.10.feed_forward1.linear2",
159
+ "encoder.layers.10.self_attn.q_proj",
160
+ "encoder.layers.10.self_attn.k_proj",
161
+ "encoder.layers.10.self_attn.v_proj",
162
+ "encoder.layers.10.self_attn.o_proj",
163
+ "encoder.layers.10.self_attn.relative_k_proj",
164
+ "encoder.layers.10.conv.pointwise_conv1",
165
+ "encoder.layers.10.conv.pointwise_conv2",
166
+ "encoder.layers.10.feed_forward2.linear1",
167
+ "encoder.layers.10.feed_forward2.linear2",
168
+ "encoder.layers.11.feed_forward1.linear1",
169
+ "encoder.layers.11.feed_forward1.linear2",
170
+ "encoder.layers.11.self_attn.q_proj",
171
+ "encoder.layers.11.self_attn.k_proj",
172
+ "encoder.layers.11.self_attn.v_proj",
173
+ "encoder.layers.11.self_attn.o_proj",
174
+ "encoder.layers.11.self_attn.relative_k_proj",
175
+ "encoder.layers.11.conv.pointwise_conv1",
176
+ "encoder.layers.11.conv.pointwise_conv2",
177
+ "encoder.layers.11.feed_forward2.linear1",
178
+ "encoder.layers.11.feed_forward2.linear2",
179
+ "encoder.layers.12.feed_forward1.linear1",
180
+ "encoder.layers.12.feed_forward1.linear2",
181
+ "encoder.layers.12.self_attn.q_proj",
182
+ "encoder.layers.12.self_attn.k_proj",
183
+ "encoder.layers.12.self_attn.v_proj",
184
+ "encoder.layers.12.self_attn.o_proj",
185
+ "encoder.layers.12.self_attn.relative_k_proj",
186
+ "encoder.layers.12.conv.pointwise_conv1",
187
+ "encoder.layers.12.conv.pointwise_conv2",
188
+ "encoder.layers.12.feed_forward2.linear1",
189
+ "encoder.layers.12.feed_forward2.linear2",
190
+ "encoder.layers.13.feed_forward1.linear1",
191
+ "encoder.layers.13.feed_forward1.linear2",
192
+ "encoder.layers.13.self_attn.q_proj",
193
+ "encoder.layers.13.self_attn.k_proj",
194
+ "encoder.layers.13.self_attn.v_proj",
195
+ "encoder.layers.13.self_attn.o_proj",
196
+ "encoder.layers.13.self_attn.relative_k_proj",
197
+ "encoder.layers.13.conv.pointwise_conv1",
198
+ "encoder.layers.13.conv.pointwise_conv2",
199
+ "encoder.layers.13.feed_forward2.linear1",
200
+ "encoder.layers.13.feed_forward2.linear2",
201
+ "encoder.layers.14.feed_forward1.linear1",
202
+ "encoder.layers.14.feed_forward1.linear2",
203
+ "encoder.layers.14.self_attn.q_proj",
204
+ "encoder.layers.14.self_attn.k_proj",
205
+ "encoder.layers.14.self_attn.v_proj",
206
+ "encoder.layers.14.self_attn.o_proj",
207
+ "encoder.layers.14.self_attn.relative_k_proj",
208
+ "encoder.layers.14.conv.pointwise_conv1",
209
+ "encoder.layers.14.conv.pointwise_conv2",
210
+ "encoder.layers.14.feed_forward2.linear1",
211
+ "encoder.layers.14.feed_forward2.linear2",
212
+ "encoder.layers.15.feed_forward1.linear1",
213
+ "encoder.layers.15.feed_forward1.linear2",
214
+ "encoder.layers.15.self_attn.q_proj",
215
+ "encoder.layers.15.self_attn.k_proj",
216
+ "encoder.layers.15.self_attn.v_proj",
217
+ "encoder.layers.15.self_attn.o_proj",
218
+ "encoder.layers.15.self_attn.relative_k_proj",
219
+ "encoder.layers.15.conv.pointwise_conv1",
220
+ "encoder.layers.15.conv.pointwise_conv2",
221
+ "encoder.layers.15.feed_forward2.linear1",
222
+ "encoder.layers.15.feed_forward2.linear2",
223
+ "encoder.layers.16.feed_forward1.linear1",
224
+ "encoder.layers.16.feed_forward1.linear2",
225
+ "encoder.layers.16.self_attn.q_proj",
226
+ "encoder.layers.16.self_attn.k_proj",
227
+ "encoder.layers.16.self_attn.v_proj",
228
+ "encoder.layers.16.self_attn.o_proj",
229
+ "encoder.layers.16.self_attn.relative_k_proj",
230
+ "encoder.layers.16.conv.pointwise_conv1",
231
+ "encoder.layers.16.conv.pointwise_conv2",
232
+ "encoder.layers.16.feed_forward2.linear1",
233
+ "encoder.layers.16.feed_forward2.linear2",
234
+ "encoder.layers.17.feed_forward1.linear1",
235
+ "encoder.layers.17.feed_forward1.linear2",
236
+ "encoder.layers.17.self_attn.q_proj",
237
+ "encoder.layers.17.self_attn.k_proj",
238
+ "encoder.layers.17.self_attn.v_proj",
239
+ "encoder.layers.17.self_attn.o_proj",
240
+ "encoder.layers.17.self_attn.relative_k_proj",
241
+ "encoder.layers.17.conv.pointwise_conv1",
242
+ "encoder.layers.17.conv.pointwise_conv2",
243
+ "encoder.layers.17.feed_forward2.linear1",
244
+ "encoder.layers.17.feed_forward2.linear2",
245
+ "encoder.layers.18.feed_forward1.linear1",
246
+ "encoder.layers.18.feed_forward1.linear2",
247
+ "encoder.layers.18.self_attn.q_proj",
248
+ "encoder.layers.18.self_attn.k_proj",
249
+ "encoder.layers.18.self_attn.v_proj",
250
+ "encoder.layers.18.self_attn.o_proj",
251
+ "encoder.layers.18.self_attn.relative_k_proj",
252
+ "encoder.layers.18.conv.pointwise_conv1",
253
+ "encoder.layers.18.conv.pointwise_conv2",
254
+ "encoder.layers.18.feed_forward2.linear1",
255
+ "encoder.layers.18.feed_forward2.linear2",
256
+ "encoder.layers.19.feed_forward1.linear1",
257
+ "encoder.layers.19.feed_forward1.linear2",
258
+ "encoder.layers.19.self_attn.q_proj",
259
+ "encoder.layers.19.self_attn.k_proj",
260
+ "encoder.layers.19.self_attn.v_proj",
261
+ "encoder.layers.19.self_attn.o_proj",
262
+ "encoder.layers.19.self_attn.relative_k_proj",
263
+ "encoder.layers.19.conv.pointwise_conv1",
264
+ "encoder.layers.19.conv.pointwise_conv2",
265
+ "encoder.layers.19.feed_forward2.linear1",
266
+ "encoder.layers.19.feed_forward2.linear2",
267
+ "encoder.layers.20.feed_forward1.linear1",
268
+ "encoder.layers.20.feed_forward1.linear2",
269
+ "encoder.layers.20.self_attn.q_proj",
270
+ "encoder.layers.20.self_attn.k_proj",
271
+ "encoder.layers.20.self_attn.v_proj",
272
+ "encoder.layers.20.self_attn.o_proj",
273
+ "encoder.layers.20.self_attn.relative_k_proj",
274
+ "encoder.layers.20.conv.pointwise_conv1",
275
+ "encoder.layers.20.conv.pointwise_conv2",
276
+ "encoder.layers.20.feed_forward2.linear1",
277
+ "encoder.layers.20.feed_forward2.linear2",
278
+ "encoder.layers.21.feed_forward1.linear1",
279
+ "encoder.layers.21.feed_forward1.linear2",
280
+ "encoder.layers.21.self_attn.q_proj",
281
+ "encoder.layers.21.self_attn.k_proj",
282
+ "encoder.layers.21.self_attn.v_proj",
283
+ "encoder.layers.21.self_attn.o_proj",
284
+ "encoder.layers.21.self_attn.relative_k_proj",
285
+ "encoder.layers.21.conv.pointwise_conv1",
286
+ "encoder.layers.21.conv.pointwise_conv2",
287
+ "encoder.layers.21.feed_forward2.linear1",
288
+ "encoder.layers.21.feed_forward2.linear2",
289
+ "encoder.layers.22.feed_forward1.linear1",
290
+ "encoder.layers.22.feed_forward1.linear2",
291
+ "encoder.layers.22.self_attn.q_proj",
292
+ "encoder.layers.22.self_attn.k_proj",
293
+ "encoder.layers.22.self_attn.v_proj",
294
+ "encoder.layers.22.self_attn.o_proj",
295
+ "encoder.layers.22.self_attn.relative_k_proj",
296
+ "encoder.layers.22.conv.pointwise_conv1",
297
+ "encoder.layers.22.conv.pointwise_conv2",
298
+ "encoder.layers.22.feed_forward2.linear1",
299
+ "encoder.layers.22.feed_forward2.linear2",
300
+ "encoder.layers.23.feed_forward1.linear1",
301
+ "encoder.layers.23.feed_forward1.linear2",
302
+ "encoder.layers.23.self_attn.q_proj",
303
+ "encoder.layers.23.self_attn.k_proj",
304
+ "encoder.layers.23.self_attn.v_proj",
305
+ "encoder.layers.23.self_attn.o_proj",
306
+ "encoder.layers.23.self_attn.relative_k_proj",
307
+ "encoder.layers.23.conv.pointwise_conv1",
308
+ "encoder.layers.23.conv.pointwise_conv2",
309
+ "encoder.layers.23.feed_forward2.linear1",
310
+ "encoder.layers.23.feed_forward2.linear2"
311
+ ]
312
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:78ec25733ee0d0c1586d1346fc86db9d0c2e436e3a8ab1d32a82d1bb8f848d21
3
+ size 177774490
ternary.json ADDED
@@ -0,0 +1,2417 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "format": "thrush-ternary-v2",
3
+ "names": "hf",
4
+ "packing": {
5
+ "base": 3,
6
+ "code_offset": 1,
7
+ "elements_per_byte": 5,
8
+ "row_bytes": "ceil(in/5)",
9
+ "digit_order": "element i of a row is base-3 digit i%5 of byte i//5, least significant digit first; pad digits 0",
10
+ "row_major": true
11
+ },
12
+ "weight_rule": "w = scales[row, col // group_size] * (code - 1) with code in {0,1,2}",
13
+ "quant": {
14
+ "mode": "ternary",
15
+ "group_size": 128
16
+ },
17
+ "source": {
18
+ "teacher": "nvidia/parakeet-tdt-0.6b-v3"
19
+ },
20
+ "vad_head": {
21
+ "tensors": [
22
+ "vad_head.ctx.bias",
23
+ "vad_head.ctx.weight",
24
+ "vad_head.out.bias",
25
+ "vad_head.out.weight",
26
+ "vad_head.proj.bias",
27
+ "vad_head.proj.weight"
28
+ ],
29
+ "d_model": 1024,
30
+ "hidden": 128,
31
+ "kernel": 5,
32
+ "frame_seconds": 0.08,
33
+ "dtype": "float16",
34
+ "params": 213377
35
+ },
36
+ "quantized_modules": [
37
+ {
38
+ "name": "encoder.layers.0.feed_forward1.linear1",
39
+ "out_features": 4096,
40
+ "in_features": 1024,
41
+ "group_size": 128,
42
+ "as_conv1d": false,
43
+ "has_bias": false,
44
+ "zero_fraction": 0.5044150352478027
45
+ },
46
+ {
47
+ "name": "encoder.layers.0.feed_forward1.linear2",
48
+ "out_features": 1024,
49
+ "in_features": 4096,
50
+ "group_size": 128,
51
+ "as_conv1d": false,
52
+ "has_bias": false,
53
+ "zero_fraction": 0.5184123516082764
54
+ },
55
+ {
56
+ "name": "encoder.layers.0.self_attn.q_proj",
57
+ "out_features": 1024,
58
+ "in_features": 1024,
59
+ "group_size": 128,
60
+ "as_conv1d": false,
61
+ "has_bias": false,
62
+ "zero_fraction": 0.15679550170898438
63
+ },
64
+ {
65
+ "name": "encoder.layers.0.self_attn.k_proj",
66
+ "out_features": 1024,
67
+ "in_features": 1024,
68
+ "group_size": 128,
69
+ "as_conv1d": false,
70
+ "has_bias": false,
71
+ "zero_fraction": 0.46837711334228516
72
+ },
73
+ {
74
+ "name": "encoder.layers.0.self_attn.v_proj",
75
+ "out_features": 1024,
76
+ "in_features": 1024,
77
+ "group_size": 128,
78
+ "as_conv1d": false,
79
+ "has_bias": false,
80
+ "zero_fraction": 0.7815647125244141
81
+ },
82
+ {
83
+ "name": "encoder.layers.0.self_attn.o_proj",
84
+ "out_features": 1024,
85
+ "in_features": 1024,
86
+ "group_size": 128,
87
+ "as_conv1d": false,
88
+ "has_bias": false,
89
+ "zero_fraction": 0.7256069183349609
90
+ },
91
+ {
92
+ "name": "encoder.layers.0.self_attn.relative_k_proj",
93
+ "out_features": 1024,
94
+ "in_features": 1024,
95
+ "group_size": 128,
96
+ "as_conv1d": false,
97
+ "has_bias": false,
98
+ "zero_fraction": 0.13124942779541016
99
+ },
100
+ {
101
+ "name": "encoder.layers.0.conv.pointwise_conv1",
102
+ "out_features": 2048,
103
+ "in_features": 1024,
104
+ "group_size": 128,
105
+ "as_conv1d": true,
106
+ "has_bias": false,
107
+ "zero_fraction": 0.49237823486328125
108
+ },
109
+ {
110
+ "name": "encoder.layers.0.conv.pointwise_conv2",
111
+ "out_features": 1024,
112
+ "in_features": 1024,
113
+ "group_size": 128,
114
+ "as_conv1d": true,
115
+ "has_bias": false,
116
+ "zero_fraction": 0.5364360809326172
117
+ },
118
+ {
119
+ "name": "encoder.layers.0.feed_forward2.linear1",
120
+ "out_features": 4096,
121
+ "in_features": 1024,
122
+ "group_size": 128,
123
+ "as_conv1d": false,
124
+ "has_bias": false,
125
+ "zero_fraction": 0.5409379005432129
126
+ },
127
+ {
128
+ "name": "encoder.layers.0.feed_forward2.linear2",
129
+ "out_features": 1024,
130
+ "in_features": 4096,
131
+ "group_size": 128,
132
+ "as_conv1d": false,
133
+ "has_bias": false,
134
+ "zero_fraction": 0.5457351207733154
135
+ },
136
+ {
137
+ "name": "encoder.layers.1.feed_forward1.linear1",
138
+ "out_features": 4096,
139
+ "in_features": 1024,
140
+ "group_size": 128,
141
+ "as_conv1d": false,
142
+ "has_bias": false,
143
+ "zero_fraction": 0.5231430530548096
144
+ },
145
+ {
146
+ "name": "encoder.layers.1.feed_forward1.linear2",
147
+ "out_features": 1024,
148
+ "in_features": 4096,
149
+ "group_size": 128,
150
+ "as_conv1d": false,
151
+ "has_bias": false,
152
+ "zero_fraction": 0.4283597469329834
153
+ },
154
+ {
155
+ "name": "encoder.layers.1.self_attn.q_proj",
156
+ "out_features": 1024,
157
+ "in_features": 1024,
158
+ "group_size": 128,
159
+ "as_conv1d": false,
160
+ "has_bias": false,
161
+ "zero_fraction": 0.34552860260009766
162
+ },
163
+ {
164
+ "name": "encoder.layers.1.self_attn.k_proj",
165
+ "out_features": 1024,
166
+ "in_features": 1024,
167
+ "group_size": 128,
168
+ "as_conv1d": false,
169
+ "has_bias": false,
170
+ "zero_fraction": 0.5812540054321289
171
+ },
172
+ {
173
+ "name": "encoder.layers.1.self_attn.v_proj",
174
+ "out_features": 1024,
175
+ "in_features": 1024,
176
+ "group_size": 128,
177
+ "as_conv1d": false,
178
+ "has_bias": false,
179
+ "zero_fraction": 0.41185569763183594
180
+ },
181
+ {
182
+ "name": "encoder.layers.1.self_attn.o_proj",
183
+ "out_features": 1024,
184
+ "in_features": 1024,
185
+ "group_size": 128,
186
+ "as_conv1d": false,
187
+ "has_bias": false,
188
+ "zero_fraction": 0.3538980484008789
189
+ },
190
+ {
191
+ "name": "encoder.layers.1.self_attn.relative_k_proj",
192
+ "out_features": 1024,
193
+ "in_features": 1024,
194
+ "group_size": 128,
195
+ "as_conv1d": false,
196
+ "has_bias": false,
197
+ "zero_fraction": 0.13209915161132812
198
+ },
199
+ {
200
+ "name": "encoder.layers.1.conv.pointwise_conv1",
201
+ "out_features": 2048,
202
+ "in_features": 1024,
203
+ "group_size": 128,
204
+ "as_conv1d": true,
205
+ "has_bias": false,
206
+ "zero_fraction": 0.4484829902648926
207
+ },
208
+ {
209
+ "name": "encoder.layers.1.conv.pointwise_conv2",
210
+ "out_features": 1024,
211
+ "in_features": 1024,
212
+ "group_size": 128,
213
+ "as_conv1d": true,
214
+ "has_bias": false,
215
+ "zero_fraction": 0.43465423583984375
216
+ },
217
+ {
218
+ "name": "encoder.layers.1.feed_forward2.linear1",
219
+ "out_features": 4096,
220
+ "in_features": 1024,
221
+ "group_size": 128,
222
+ "as_conv1d": false,
223
+ "has_bias": false,
224
+ "zero_fraction": 0.42896461486816406
225
+ },
226
+ {
227
+ "name": "encoder.layers.1.feed_forward2.linear2",
228
+ "out_features": 1024,
229
+ "in_features": 4096,
230
+ "group_size": 128,
231
+ "as_conv1d": false,
232
+ "has_bias": false,
233
+ "zero_fraction": 0.5293712615966797
234
+ },
235
+ {
236
+ "name": "encoder.layers.2.feed_forward1.linear1",
237
+ "out_features": 4096,
238
+ "in_features": 1024,
239
+ "group_size": 128,
240
+ "as_conv1d": false,
241
+ "has_bias": false,
242
+ "zero_fraction": 0.4617030620574951
243
+ },
244
+ {
245
+ "name": "encoder.layers.2.feed_forward1.linear2",
246
+ "out_features": 1024,
247
+ "in_features": 4096,
248
+ "group_size": 128,
249
+ "as_conv1d": false,
250
+ "has_bias": false,
251
+ "zero_fraction": 0.45351243019104004
252
+ },
253
+ {
254
+ "name": "encoder.layers.2.self_attn.q_proj",
255
+ "out_features": 1024,
256
+ "in_features": 1024,
257
+ "group_size": 128,
258
+ "as_conv1d": false,
259
+ "has_bias": false,
260
+ "zero_fraction": 0.32970237731933594
261
+ },
262
+ {
263
+ "name": "encoder.layers.2.self_attn.k_proj",
264
+ "out_features": 1024,
265
+ "in_features": 1024,
266
+ "group_size": 128,
267
+ "as_conv1d": false,
268
+ "has_bias": false,
269
+ "zero_fraction": 0.5406455993652344
270
+ },
271
+ {
272
+ "name": "encoder.layers.2.self_attn.v_proj",
273
+ "out_features": 1024,
274
+ "in_features": 1024,
275
+ "group_size": 128,
276
+ "as_conv1d": false,
277
+ "has_bias": false,
278
+ "zero_fraction": 0.40648365020751953
279
+ },
280
+ {
281
+ "name": "encoder.layers.2.self_attn.o_proj",
282
+ "out_features": 1024,
283
+ "in_features": 1024,
284
+ "group_size": 128,
285
+ "as_conv1d": false,
286
+ "has_bias": false,
287
+ "zero_fraction": 0.38976478576660156
288
+ },
289
+ {
290
+ "name": "encoder.layers.2.self_attn.relative_k_proj",
291
+ "out_features": 1024,
292
+ "in_features": 1024,
293
+ "group_size": 128,
294
+ "as_conv1d": false,
295
+ "has_bias": false,
296
+ "zero_fraction": 0.1539907455444336
297
+ },
298
+ {
299
+ "name": "encoder.layers.2.conv.pointwise_conv1",
300
+ "out_features": 2048,
301
+ "in_features": 1024,
302
+ "group_size": 128,
303
+ "as_conv1d": true,
304
+ "has_bias": false,
305
+ "zero_fraction": 0.45189952850341797
306
+ },
307
+ {
308
+ "name": "encoder.layers.2.conv.pointwise_conv2",
309
+ "out_features": 1024,
310
+ "in_features": 1024,
311
+ "group_size": 128,
312
+ "as_conv1d": true,
313
+ "has_bias": false,
314
+ "zero_fraction": 0.45950984954833984
315
+ },
316
+ {
317
+ "name": "encoder.layers.2.feed_forward2.linear1",
318
+ "out_features": 4096,
319
+ "in_features": 1024,
320
+ "group_size": 128,
321
+ "as_conv1d": false,
322
+ "has_bias": false,
323
+ "zero_fraction": 0.43088412284851074
324
+ },
325
+ {
326
+ "name": "encoder.layers.2.feed_forward2.linear2",
327
+ "out_features": 1024,
328
+ "in_features": 4096,
329
+ "group_size": 128,
330
+ "as_conv1d": false,
331
+ "has_bias": false,
332
+ "zero_fraction": 0.5336544513702393
333
+ },
334
+ {
335
+ "name": "encoder.layers.3.feed_forward1.linear1",
336
+ "out_features": 4096,
337
+ "in_features": 1024,
338
+ "group_size": 128,
339
+ "as_conv1d": false,
340
+ "has_bias": false,
341
+ "zero_fraction": 0.4594447612762451
342
+ },
343
+ {
344
+ "name": "encoder.layers.3.feed_forward1.linear2",
345
+ "out_features": 1024,
346
+ "in_features": 4096,
347
+ "group_size": 128,
348
+ "as_conv1d": false,
349
+ "has_bias": false,
350
+ "zero_fraction": 0.4711878299713135
351
+ },
352
+ {
353
+ "name": "encoder.layers.3.self_attn.q_proj",
354
+ "out_features": 1024,
355
+ "in_features": 1024,
356
+ "group_size": 128,
357
+ "as_conv1d": false,
358
+ "has_bias": false,
359
+ "zero_fraction": 0.34746265411376953
360
+ },
361
+ {
362
+ "name": "encoder.layers.3.self_attn.k_proj",
363
+ "out_features": 1024,
364
+ "in_features": 1024,
365
+ "group_size": 128,
366
+ "as_conv1d": false,
367
+ "has_bias": false,
368
+ "zero_fraction": 0.5459375381469727
369
+ },
370
+ {
371
+ "name": "encoder.layers.3.self_attn.v_proj",
372
+ "out_features": 1024,
373
+ "in_features": 1024,
374
+ "group_size": 128,
375
+ "as_conv1d": false,
376
+ "has_bias": false,
377
+ "zero_fraction": 0.3894500732421875
378
+ },
379
+ {
380
+ "name": "encoder.layers.3.self_attn.o_proj",
381
+ "out_features": 1024,
382
+ "in_features": 1024,
383
+ "group_size": 128,
384
+ "as_conv1d": false,
385
+ "has_bias": false,
386
+ "zero_fraction": 0.37818431854248047
387
+ },
388
+ {
389
+ "name": "encoder.layers.3.self_attn.relative_k_proj",
390
+ "out_features": 1024,
391
+ "in_features": 1024,
392
+ "group_size": 128,
393
+ "as_conv1d": false,
394
+ "has_bias": false,
395
+ "zero_fraction": 0.15802383422851562
396
+ },
397
+ {
398
+ "name": "encoder.layers.3.conv.pointwise_conv1",
399
+ "out_features": 2048,
400
+ "in_features": 1024,
401
+ "group_size": 128,
402
+ "as_conv1d": true,
403
+ "has_bias": false,
404
+ "zero_fraction": 0.4486575126647949
405
+ },
406
+ {
407
+ "name": "encoder.layers.3.conv.pointwise_conv2",
408
+ "out_features": 1024,
409
+ "in_features": 1024,
410
+ "group_size": 128,
411
+ "as_conv1d": true,
412
+ "has_bias": false,
413
+ "zero_fraction": 0.4589385986328125
414
+ },
415
+ {
416
+ "name": "encoder.layers.3.feed_forward2.linear1",
417
+ "out_features": 4096,
418
+ "in_features": 1024,
419
+ "group_size": 128,
420
+ "as_conv1d": false,
421
+ "has_bias": false,
422
+ "zero_fraction": 0.4405934810638428
423
+ },
424
+ {
425
+ "name": "encoder.layers.3.feed_forward2.linear2",
426
+ "out_features": 1024,
427
+ "in_features": 4096,
428
+ "group_size": 128,
429
+ "as_conv1d": false,
430
+ "has_bias": false,
431
+ "zero_fraction": 0.540421724319458
432
+ },
433
+ {
434
+ "name": "encoder.layers.4.feed_forward1.linear1",
435
+ "out_features": 4096,
436
+ "in_features": 1024,
437
+ "group_size": 128,
438
+ "as_conv1d": false,
439
+ "has_bias": false,
440
+ "zero_fraction": 0.4638183116912842
441
+ },
442
+ {
443
+ "name": "encoder.layers.4.feed_forward1.linear2",
444
+ "out_features": 1024,
445
+ "in_features": 4096,
446
+ "group_size": 128,
447
+ "as_conv1d": false,
448
+ "has_bias": false,
449
+ "zero_fraction": 0.48339200019836426
450
+ },
451
+ {
452
+ "name": "encoder.layers.4.self_attn.q_proj",
453
+ "out_features": 1024,
454
+ "in_features": 1024,
455
+ "group_size": 128,
456
+ "as_conv1d": false,
457
+ "has_bias": false,
458
+ "zero_fraction": 0.3684682846069336
459
+ },
460
+ {
461
+ "name": "encoder.layers.4.self_attn.k_proj",
462
+ "out_features": 1024,
463
+ "in_features": 1024,
464
+ "group_size": 128,
465
+ "as_conv1d": false,
466
+ "has_bias": false,
467
+ "zero_fraction": 0.5612239837646484
468
+ },
469
+ {
470
+ "name": "encoder.layers.4.self_attn.v_proj",
471
+ "out_features": 1024,
472
+ "in_features": 1024,
473
+ "group_size": 128,
474
+ "as_conv1d": false,
475
+ "has_bias": false,
476
+ "zero_fraction": 0.38426876068115234
477
+ },
478
+ {
479
+ "name": "encoder.layers.4.self_attn.o_proj",
480
+ "out_features": 1024,
481
+ "in_features": 1024,
482
+ "group_size": 128,
483
+ "as_conv1d": false,
484
+ "has_bias": false,
485
+ "zero_fraction": 0.3797454833984375
486
+ },
487
+ {
488
+ "name": "encoder.layers.4.self_attn.relative_k_proj",
489
+ "out_features": 1024,
490
+ "in_features": 1024,
491
+ "group_size": 128,
492
+ "as_conv1d": false,
493
+ "has_bias": false,
494
+ "zero_fraction": 0.14944171905517578
495
+ },
496
+ {
497
+ "name": "encoder.layers.4.conv.pointwise_conv1",
498
+ "out_features": 2048,
499
+ "in_features": 1024,
500
+ "group_size": 128,
501
+ "as_conv1d": true,
502
+ "has_bias": false,
503
+ "zero_fraction": 0.45036983489990234
504
+ },
505
+ {
506
+ "name": "encoder.layers.4.conv.pointwise_conv2",
507
+ "out_features": 1024,
508
+ "in_features": 1024,
509
+ "group_size": 128,
510
+ "as_conv1d": true,
511
+ "has_bias": false,
512
+ "zero_fraction": 0.4643983840942383
513
+ },
514
+ {
515
+ "name": "encoder.layers.4.feed_forward2.linear1",
516
+ "out_features": 4096,
517
+ "in_features": 1024,
518
+ "group_size": 128,
519
+ "as_conv1d": false,
520
+ "has_bias": false,
521
+ "zero_fraction": 0.447796106338501
522
+ },
523
+ {
524
+ "name": "encoder.layers.4.feed_forward2.linear2",
525
+ "out_features": 1024,
526
+ "in_features": 4096,
527
+ "group_size": 128,
528
+ "as_conv1d": false,
529
+ "has_bias": false,
530
+ "zero_fraction": 0.547126054763794
531
+ },
532
+ {
533
+ "name": "encoder.layers.5.feed_forward1.linear1",
534
+ "out_features": 4096,
535
+ "in_features": 1024,
536
+ "group_size": 128,
537
+ "as_conv1d": false,
538
+ "has_bias": false,
539
+ "zero_fraction": 0.47231388092041016
540
+ },
541
+ {
542
+ "name": "encoder.layers.5.feed_forward1.linear2",
543
+ "out_features": 1024,
544
+ "in_features": 4096,
545
+ "group_size": 128,
546
+ "as_conv1d": false,
547
+ "has_bias": false,
548
+ "zero_fraction": 0.48183250427246094
549
+ },
550
+ {
551
+ "name": "encoder.layers.5.self_attn.q_proj",
552
+ "out_features": 1024,
553
+ "in_features": 1024,
554
+ "group_size": 128,
555
+ "as_conv1d": false,
556
+ "has_bias": false,
557
+ "zero_fraction": 0.3488779067993164
558
+ },
559
+ {
560
+ "name": "encoder.layers.5.self_attn.k_proj",
561
+ "out_features": 1024,
562
+ "in_features": 1024,
563
+ "group_size": 128,
564
+ "as_conv1d": false,
565
+ "has_bias": false,
566
+ "zero_fraction": 0.5331039428710938
567
+ },
568
+ {
569
+ "name": "encoder.layers.5.self_attn.v_proj",
570
+ "out_features": 1024,
571
+ "in_features": 1024,
572
+ "group_size": 128,
573
+ "as_conv1d": false,
574
+ "has_bias": false,
575
+ "zero_fraction": 0.40555763244628906
576
+ },
577
+ {
578
+ "name": "encoder.layers.5.self_attn.o_proj",
579
+ "out_features": 1024,
580
+ "in_features": 1024,
581
+ "group_size": 128,
582
+ "as_conv1d": false,
583
+ "has_bias": false,
584
+ "zero_fraction": 0.3914680480957031
585
+ },
586
+ {
587
+ "name": "encoder.layers.5.self_attn.relative_k_proj",
588
+ "out_features": 1024,
589
+ "in_features": 1024,
590
+ "group_size": 128,
591
+ "as_conv1d": false,
592
+ "has_bias": false,
593
+ "zero_fraction": 0.15366840362548828
594
+ },
595
+ {
596
+ "name": "encoder.layers.5.conv.pointwise_conv1",
597
+ "out_features": 2048,
598
+ "in_features": 1024,
599
+ "group_size": 128,
600
+ "as_conv1d": true,
601
+ "has_bias": false,
602
+ "zero_fraction": 0.45035457611083984
603
+ },
604
+ {
605
+ "name": "encoder.layers.5.conv.pointwise_conv2",
606
+ "out_features": 1024,
607
+ "in_features": 1024,
608
+ "group_size": 128,
609
+ "as_conv1d": true,
610
+ "has_bias": false,
611
+ "zero_fraction": 0.4670143127441406
612
+ },
613
+ {
614
+ "name": "encoder.layers.5.feed_forward2.linear1",
615
+ "out_features": 4096,
616
+ "in_features": 1024,
617
+ "group_size": 128,
618
+ "as_conv1d": false,
619
+ "has_bias": false,
620
+ "zero_fraction": 0.45580148696899414
621
+ },
622
+ {
623
+ "name": "encoder.layers.5.feed_forward2.linear2",
624
+ "out_features": 1024,
625
+ "in_features": 4096,
626
+ "group_size": 128,
627
+ "as_conv1d": false,
628
+ "has_bias": false,
629
+ "zero_fraction": 0.5230386257171631
630
+ },
631
+ {
632
+ "name": "encoder.layers.6.feed_forward1.linear1",
633
+ "out_features": 4096,
634
+ "in_features": 1024,
635
+ "group_size": 128,
636
+ "as_conv1d": false,
637
+ "has_bias": false,
638
+ "zero_fraction": 0.4766519069671631
639
+ },
640
+ {
641
+ "name": "encoder.layers.6.feed_forward1.linear2",
642
+ "out_features": 1024,
643
+ "in_features": 4096,
644
+ "group_size": 128,
645
+ "as_conv1d": false,
646
+ "has_bias": false,
647
+ "zero_fraction": 0.4740920066833496
648
+ },
649
+ {
650
+ "name": "encoder.layers.6.self_attn.q_proj",
651
+ "out_features": 1024,
652
+ "in_features": 1024,
653
+ "group_size": 128,
654
+ "as_conv1d": false,
655
+ "has_bias": false,
656
+ "zero_fraction": 0.3270740509033203
657
+ },
658
+ {
659
+ "name": "encoder.layers.6.self_attn.k_proj",
660
+ "out_features": 1024,
661
+ "in_features": 1024,
662
+ "group_size": 128,
663
+ "as_conv1d": false,
664
+ "has_bias": false,
665
+ "zero_fraction": 0.5052204132080078
666
+ },
667
+ {
668
+ "name": "encoder.layers.6.self_attn.v_proj",
669
+ "out_features": 1024,
670
+ "in_features": 1024,
671
+ "group_size": 128,
672
+ "as_conv1d": false,
673
+ "has_bias": false,
674
+ "zero_fraction": 0.4151468276977539
675
+ },
676
+ {
677
+ "name": "encoder.layers.6.self_attn.o_proj",
678
+ "out_features": 1024,
679
+ "in_features": 1024,
680
+ "group_size": 128,
681
+ "as_conv1d": false,
682
+ "has_bias": false,
683
+ "zero_fraction": 0.4025430679321289
684
+ },
685
+ {
686
+ "name": "encoder.layers.6.self_attn.relative_k_proj",
687
+ "out_features": 1024,
688
+ "in_features": 1024,
689
+ "group_size": 128,
690
+ "as_conv1d": false,
691
+ "has_bias": false,
692
+ "zero_fraction": 0.15303707122802734
693
+ },
694
+ {
695
+ "name": "encoder.layers.6.conv.pointwise_conv1",
696
+ "out_features": 2048,
697
+ "in_features": 1024,
698
+ "group_size": 128,
699
+ "as_conv1d": true,
700
+ "has_bias": false,
701
+ "zero_fraction": 0.45003604888916016
702
+ },
703
+ {
704
+ "name": "encoder.layers.6.conv.pointwise_conv2",
705
+ "out_features": 1024,
706
+ "in_features": 1024,
707
+ "group_size": 128,
708
+ "as_conv1d": true,
709
+ "has_bias": false,
710
+ "zero_fraction": 0.4495658874511719
711
+ },
712
+ {
713
+ "name": "encoder.layers.6.feed_forward2.linear1",
714
+ "out_features": 4096,
715
+ "in_features": 1024,
716
+ "group_size": 128,
717
+ "as_conv1d": false,
718
+ "has_bias": false,
719
+ "zero_fraction": 0.4526941776275635
720
+ },
721
+ {
722
+ "name": "encoder.layers.6.feed_forward2.linear2",
723
+ "out_features": 1024,
724
+ "in_features": 4096,
725
+ "group_size": 128,
726
+ "as_conv1d": false,
727
+ "has_bias": false,
728
+ "zero_fraction": 0.5148673057556152
729
+ },
730
+ {
731
+ "name": "encoder.layers.7.feed_forward1.linear1",
732
+ "out_features": 4096,
733
+ "in_features": 1024,
734
+ "group_size": 128,
735
+ "as_conv1d": false,
736
+ "has_bias": false,
737
+ "zero_fraction": 0.48210573196411133
738
+ },
739
+ {
740
+ "name": "encoder.layers.7.feed_forward1.linear2",
741
+ "out_features": 1024,
742
+ "in_features": 4096,
743
+ "group_size": 128,
744
+ "as_conv1d": false,
745
+ "has_bias": false,
746
+ "zero_fraction": 0.46149134635925293
747
+ },
748
+ {
749
+ "name": "encoder.layers.7.self_attn.q_proj",
750
+ "out_features": 1024,
751
+ "in_features": 1024,
752
+ "group_size": 128,
753
+ "as_conv1d": false,
754
+ "has_bias": false,
755
+ "zero_fraction": 0.33362388610839844
756
+ },
757
+ {
758
+ "name": "encoder.layers.7.self_attn.k_proj",
759
+ "out_features": 1024,
760
+ "in_features": 1024,
761
+ "group_size": 128,
762
+ "as_conv1d": false,
763
+ "has_bias": false,
764
+ "zero_fraction": 0.5477542877197266
765
+ },
766
+ {
767
+ "name": "encoder.layers.7.self_attn.v_proj",
768
+ "out_features": 1024,
769
+ "in_features": 1024,
770
+ "group_size": 128,
771
+ "as_conv1d": false,
772
+ "has_bias": false,
773
+ "zero_fraction": 0.42502307891845703
774
+ },
775
+ {
776
+ "name": "encoder.layers.7.self_attn.o_proj",
777
+ "out_features": 1024,
778
+ "in_features": 1024,
779
+ "group_size": 128,
780
+ "as_conv1d": false,
781
+ "has_bias": false,
782
+ "zero_fraction": 0.3920621871948242
783
+ },
784
+ {
785
+ "name": "encoder.layers.7.self_attn.relative_k_proj",
786
+ "out_features": 1024,
787
+ "in_features": 1024,
788
+ "group_size": 128,
789
+ "as_conv1d": false,
790
+ "has_bias": false,
791
+ "zero_fraction": 0.1467151641845703
792
+ },
793
+ {
794
+ "name": "encoder.layers.7.conv.pointwise_conv1",
795
+ "out_features": 2048,
796
+ "in_features": 1024,
797
+ "group_size": 128,
798
+ "as_conv1d": true,
799
+ "has_bias": false,
800
+ "zero_fraction": 0.4432258605957031
801
+ },
802
+ {
803
+ "name": "encoder.layers.7.conv.pointwise_conv2",
804
+ "out_features": 1024,
805
+ "in_features": 1024,
806
+ "group_size": 128,
807
+ "as_conv1d": true,
808
+ "has_bias": false,
809
+ "zero_fraction": 0.4387655258178711
810
+ },
811
+ {
812
+ "name": "encoder.layers.7.feed_forward2.linear1",
813
+ "out_features": 4096,
814
+ "in_features": 1024,
815
+ "group_size": 128,
816
+ "as_conv1d": false,
817
+ "has_bias": false,
818
+ "zero_fraction": 0.4537210464477539
819
+ },
820
+ {
821
+ "name": "encoder.layers.7.feed_forward2.linear2",
822
+ "out_features": 1024,
823
+ "in_features": 4096,
824
+ "group_size": 128,
825
+ "as_conv1d": false,
826
+ "has_bias": false,
827
+ "zero_fraction": 0.5217123031616211
828
+ },
829
+ {
830
+ "name": "encoder.layers.8.feed_forward1.linear1",
831
+ "out_features": 4096,
832
+ "in_features": 1024,
833
+ "group_size": 128,
834
+ "as_conv1d": false,
835
+ "has_bias": false,
836
+ "zero_fraction": 0.48615384101867676
837
+ },
838
+ {
839
+ "name": "encoder.layers.8.feed_forward1.linear2",
840
+ "out_features": 1024,
841
+ "in_features": 4096,
842
+ "group_size": 128,
843
+ "as_conv1d": false,
844
+ "has_bias": false,
845
+ "zero_fraction": 0.4634566307067871
846
+ },
847
+ {
848
+ "name": "encoder.layers.8.self_attn.q_proj",
849
+ "out_features": 1024,
850
+ "in_features": 1024,
851
+ "group_size": 128,
852
+ "as_conv1d": false,
853
+ "has_bias": false,
854
+ "zero_fraction": 0.34348583221435547
855
+ },
856
+ {
857
+ "name": "encoder.layers.8.self_attn.k_proj",
858
+ "out_features": 1024,
859
+ "in_features": 1024,
860
+ "group_size": 128,
861
+ "as_conv1d": false,
862
+ "has_bias": false,
863
+ "zero_fraction": 0.5461740493774414
864
+ },
865
+ {
866
+ "name": "encoder.layers.8.self_attn.v_proj",
867
+ "out_features": 1024,
868
+ "in_features": 1024,
869
+ "group_size": 128,
870
+ "as_conv1d": false,
871
+ "has_bias": false,
872
+ "zero_fraction": 0.3690071105957031
873
+ },
874
+ {
875
+ "name": "encoder.layers.8.self_attn.o_proj",
876
+ "out_features": 1024,
877
+ "in_features": 1024,
878
+ "group_size": 128,
879
+ "as_conv1d": false,
880
+ "has_bias": false,
881
+ "zero_fraction": 0.37335777282714844
882
+ },
883
+ {
884
+ "name": "encoder.layers.8.self_attn.relative_k_proj",
885
+ "out_features": 1024,
886
+ "in_features": 1024,
887
+ "group_size": 128,
888
+ "as_conv1d": false,
889
+ "has_bias": false,
890
+ "zero_fraction": 0.14555931091308594
891
+ },
892
+ {
893
+ "name": "encoder.layers.8.conv.pointwise_conv1",
894
+ "out_features": 2048,
895
+ "in_features": 1024,
896
+ "group_size": 128,
897
+ "as_conv1d": true,
898
+ "has_bias": false,
899
+ "zero_fraction": 0.4336113929748535
900
+ },
901
+ {
902
+ "name": "encoder.layers.8.conv.pointwise_conv2",
903
+ "out_features": 1024,
904
+ "in_features": 1024,
905
+ "group_size": 128,
906
+ "as_conv1d": true,
907
+ "has_bias": false,
908
+ "zero_fraction": 0.42722225189208984
909
+ },
910
+ {
911
+ "name": "encoder.layers.8.feed_forward2.linear1",
912
+ "out_features": 4096,
913
+ "in_features": 1024,
914
+ "group_size": 128,
915
+ "as_conv1d": false,
916
+ "has_bias": false,
917
+ "zero_fraction": 0.46222662925720215
918
+ },
919
+ {
920
+ "name": "encoder.layers.8.feed_forward2.linear2",
921
+ "out_features": 1024,
922
+ "in_features": 4096,
923
+ "group_size": 128,
924
+ "as_conv1d": false,
925
+ "has_bias": false,
926
+ "zero_fraction": 0.5234966278076172
927
+ },
928
+ {
929
+ "name": "encoder.layers.9.feed_forward1.linear1",
930
+ "out_features": 4096,
931
+ "in_features": 1024,
932
+ "group_size": 128,
933
+ "as_conv1d": false,
934
+ "has_bias": false,
935
+ "zero_fraction": 0.49318695068359375
936
+ },
937
+ {
938
+ "name": "encoder.layers.9.feed_forward1.linear2",
939
+ "out_features": 1024,
940
+ "in_features": 4096,
941
+ "group_size": 128,
942
+ "as_conv1d": false,
943
+ "has_bias": false,
944
+ "zero_fraction": 0.471712589263916
945
+ },
946
+ {
947
+ "name": "encoder.layers.9.self_attn.q_proj",
948
+ "out_features": 1024,
949
+ "in_features": 1024,
950
+ "group_size": 128,
951
+ "as_conv1d": false,
952
+ "has_bias": false,
953
+ "zero_fraction": 0.3633537292480469
954
+ },
955
+ {
956
+ "name": "encoder.layers.9.self_attn.k_proj",
957
+ "out_features": 1024,
958
+ "in_features": 1024,
959
+ "group_size": 128,
960
+ "as_conv1d": false,
961
+ "has_bias": false,
962
+ "zero_fraction": 0.5625839233398438
963
+ },
964
+ {
965
+ "name": "encoder.layers.9.self_attn.v_proj",
966
+ "out_features": 1024,
967
+ "in_features": 1024,
968
+ "group_size": 128,
969
+ "as_conv1d": false,
970
+ "has_bias": false,
971
+ "zero_fraction": 0.37224864959716797
972
+ },
973
+ {
974
+ "name": "encoder.layers.9.self_attn.o_proj",
975
+ "out_features": 1024,
976
+ "in_features": 1024,
977
+ "group_size": 128,
978
+ "as_conv1d": false,
979
+ "has_bias": false,
980
+ "zero_fraction": 0.3777170181274414
981
+ },
982
+ {
983
+ "name": "encoder.layers.9.self_attn.relative_k_proj",
984
+ "out_features": 1024,
985
+ "in_features": 1024,
986
+ "group_size": 128,
987
+ "as_conv1d": false,
988
+ "has_bias": false,
989
+ "zero_fraction": 0.15102577209472656
990
+ },
991
+ {
992
+ "name": "encoder.layers.9.conv.pointwise_conv1",
993
+ "out_features": 2048,
994
+ "in_features": 1024,
995
+ "group_size": 128,
996
+ "as_conv1d": true,
997
+ "has_bias": false,
998
+ "zero_fraction": 0.4328136444091797
999
+ },
1000
+ {
1001
+ "name": "encoder.layers.9.conv.pointwise_conv2",
1002
+ "out_features": 1024,
1003
+ "in_features": 1024,
1004
+ "group_size": 128,
1005
+ "as_conv1d": true,
1006
+ "has_bias": false,
1007
+ "zero_fraction": 0.41849327087402344
1008
+ },
1009
+ {
1010
+ "name": "encoder.layers.9.feed_forward2.linear1",
1011
+ "out_features": 4096,
1012
+ "in_features": 1024,
1013
+ "group_size": 128,
1014
+ "as_conv1d": false,
1015
+ "has_bias": false,
1016
+ "zero_fraction": 0.47850704193115234
1017
+ },
1018
+ {
1019
+ "name": "encoder.layers.9.feed_forward2.linear2",
1020
+ "out_features": 1024,
1021
+ "in_features": 4096,
1022
+ "group_size": 128,
1023
+ "as_conv1d": false,
1024
+ "has_bias": false,
1025
+ "zero_fraction": 0.5117299556732178
1026
+ },
1027
+ {
1028
+ "name": "encoder.layers.10.feed_forward1.linear1",
1029
+ "out_features": 4096,
1030
+ "in_features": 1024,
1031
+ "group_size": 128,
1032
+ "as_conv1d": false,
1033
+ "has_bias": false,
1034
+ "zero_fraction": 0.5326597690582275
1035
+ },
1036
+ {
1037
+ "name": "encoder.layers.10.feed_forward1.linear2",
1038
+ "out_features": 1024,
1039
+ "in_features": 4096,
1040
+ "group_size": 128,
1041
+ "as_conv1d": false,
1042
+ "has_bias": false,
1043
+ "zero_fraction": 0.4868049621582031
1044
+ },
1045
+ {
1046
+ "name": "encoder.layers.10.self_attn.q_proj",
1047
+ "out_features": 1024,
1048
+ "in_features": 1024,
1049
+ "group_size": 128,
1050
+ "as_conv1d": false,
1051
+ "has_bias": false,
1052
+ "zero_fraction": 0.38233089447021484
1053
+ },
1054
+ {
1055
+ "name": "encoder.layers.10.self_attn.k_proj",
1056
+ "out_features": 1024,
1057
+ "in_features": 1024,
1058
+ "group_size": 128,
1059
+ "as_conv1d": false,
1060
+ "has_bias": false,
1061
+ "zero_fraction": 0.6121854782104492
1062
+ },
1063
+ {
1064
+ "name": "encoder.layers.10.self_attn.v_proj",
1065
+ "out_features": 1024,
1066
+ "in_features": 1024,
1067
+ "group_size": 128,
1068
+ "as_conv1d": false,
1069
+ "has_bias": false,
1070
+ "zero_fraction": 0.3678264617919922
1071
+ },
1072
+ {
1073
+ "name": "encoder.layers.10.self_attn.o_proj",
1074
+ "out_features": 1024,
1075
+ "in_features": 1024,
1076
+ "group_size": 128,
1077
+ "as_conv1d": false,
1078
+ "has_bias": false,
1079
+ "zero_fraction": 0.36456298828125
1080
+ },
1081
+ {
1082
+ "name": "encoder.layers.10.self_attn.relative_k_proj",
1083
+ "out_features": 1024,
1084
+ "in_features": 1024,
1085
+ "group_size": 128,
1086
+ "as_conv1d": false,
1087
+ "has_bias": false,
1088
+ "zero_fraction": 0.14080429077148438
1089
+ },
1090
+ {
1091
+ "name": "encoder.layers.10.conv.pointwise_conv1",
1092
+ "out_features": 2048,
1093
+ "in_features": 1024,
1094
+ "group_size": 128,
1095
+ "as_conv1d": true,
1096
+ "has_bias": false,
1097
+ "zero_fraction": 0.4323916435241699
1098
+ },
1099
+ {
1100
+ "name": "encoder.layers.10.conv.pointwise_conv2",
1101
+ "out_features": 1024,
1102
+ "in_features": 1024,
1103
+ "group_size": 128,
1104
+ "as_conv1d": true,
1105
+ "has_bias": false,
1106
+ "zero_fraction": 0.42212963104248047
1107
+ },
1108
+ {
1109
+ "name": "encoder.layers.10.feed_forward2.linear1",
1110
+ "out_features": 4096,
1111
+ "in_features": 1024,
1112
+ "group_size": 128,
1113
+ "as_conv1d": false,
1114
+ "has_bias": false,
1115
+ "zero_fraction": 0.4855043888092041
1116
+ },
1117
+ {
1118
+ "name": "encoder.layers.10.feed_forward2.linear2",
1119
+ "out_features": 1024,
1120
+ "in_features": 4096,
1121
+ "group_size": 128,
1122
+ "as_conv1d": false,
1123
+ "has_bias": false,
1124
+ "zero_fraction": 0.5134315490722656
1125
+ },
1126
+ {
1127
+ "name": "encoder.layers.11.feed_forward1.linear1",
1128
+ "out_features": 4096,
1129
+ "in_features": 1024,
1130
+ "group_size": 128,
1131
+ "as_conv1d": false,
1132
+ "has_bias": false,
1133
+ "zero_fraction": 0.5217080116271973
1134
+ },
1135
+ {
1136
+ "name": "encoder.layers.11.feed_forward1.linear2",
1137
+ "out_features": 1024,
1138
+ "in_features": 4096,
1139
+ "group_size": 128,
1140
+ "as_conv1d": false,
1141
+ "has_bias": false,
1142
+ "zero_fraction": 0.4963808059692383
1143
+ },
1144
+ {
1145
+ "name": "encoder.layers.11.self_attn.q_proj",
1146
+ "out_features": 1024,
1147
+ "in_features": 1024,
1148
+ "group_size": 128,
1149
+ "as_conv1d": false,
1150
+ "has_bias": false,
1151
+ "zero_fraction": 0.35190773010253906
1152
+ },
1153
+ {
1154
+ "name": "encoder.layers.11.self_attn.k_proj",
1155
+ "out_features": 1024,
1156
+ "in_features": 1024,
1157
+ "group_size": 128,
1158
+ "as_conv1d": false,
1159
+ "has_bias": false,
1160
+ "zero_fraction": 0.5736579895019531
1161
+ },
1162
+ {
1163
+ "name": "encoder.layers.11.self_attn.v_proj",
1164
+ "out_features": 1024,
1165
+ "in_features": 1024,
1166
+ "group_size": 128,
1167
+ "as_conv1d": false,
1168
+ "has_bias": false,
1169
+ "zero_fraction": 0.3828563690185547
1170
+ },
1171
+ {
1172
+ "name": "encoder.layers.11.self_attn.o_proj",
1173
+ "out_features": 1024,
1174
+ "in_features": 1024,
1175
+ "group_size": 128,
1176
+ "as_conv1d": false,
1177
+ "has_bias": false,
1178
+ "zero_fraction": 0.3663148880004883
1179
+ },
1180
+ {
1181
+ "name": "encoder.layers.11.self_attn.relative_k_proj",
1182
+ "out_features": 1024,
1183
+ "in_features": 1024,
1184
+ "group_size": 128,
1185
+ "as_conv1d": false,
1186
+ "has_bias": false,
1187
+ "zero_fraction": 0.15211105346679688
1188
+ },
1189
+ {
1190
+ "name": "encoder.layers.11.conv.pointwise_conv1",
1191
+ "out_features": 2048,
1192
+ "in_features": 1024,
1193
+ "group_size": 128,
1194
+ "as_conv1d": true,
1195
+ "has_bias": false,
1196
+ "zero_fraction": 0.43471431732177734
1197
+ },
1198
+ {
1199
+ "name": "encoder.layers.11.conv.pointwise_conv2",
1200
+ "out_features": 1024,
1201
+ "in_features": 1024,
1202
+ "group_size": 128,
1203
+ "as_conv1d": true,
1204
+ "has_bias": false,
1205
+ "zero_fraction": 0.42731475830078125
1206
+ },
1207
+ {
1208
+ "name": "encoder.layers.11.feed_forward2.linear1",
1209
+ "out_features": 4096,
1210
+ "in_features": 1024,
1211
+ "group_size": 128,
1212
+ "as_conv1d": false,
1213
+ "has_bias": false,
1214
+ "zero_fraction": 0.4918859004974365
1215
+ },
1216
+ {
1217
+ "name": "encoder.layers.11.feed_forward2.linear2",
1218
+ "out_features": 1024,
1219
+ "in_features": 4096,
1220
+ "group_size": 128,
1221
+ "as_conv1d": false,
1222
+ "has_bias": false,
1223
+ "zero_fraction": 0.5454516410827637
1224
+ },
1225
+ {
1226
+ "name": "encoder.layers.12.feed_forward1.linear1",
1227
+ "out_features": 4096,
1228
+ "in_features": 1024,
1229
+ "group_size": 128,
1230
+ "as_conv1d": false,
1231
+ "has_bias": false,
1232
+ "zero_fraction": 0.5295507907867432
1233
+ },
1234
+ {
1235
+ "name": "encoder.layers.12.feed_forward1.linear2",
1236
+ "out_features": 1024,
1237
+ "in_features": 4096,
1238
+ "group_size": 128,
1239
+ "as_conv1d": false,
1240
+ "has_bias": false,
1241
+ "zero_fraction": 0.4964902400970459
1242
+ },
1243
+ {
1244
+ "name": "encoder.layers.12.self_attn.q_proj",
1245
+ "out_features": 1024,
1246
+ "in_features": 1024,
1247
+ "group_size": 128,
1248
+ "as_conv1d": false,
1249
+ "has_bias": false,
1250
+ "zero_fraction": 0.3329486846923828
1251
+ },
1252
+ {
1253
+ "name": "encoder.layers.12.self_attn.k_proj",
1254
+ "out_features": 1024,
1255
+ "in_features": 1024,
1256
+ "group_size": 128,
1257
+ "as_conv1d": false,
1258
+ "has_bias": false,
1259
+ "zero_fraction": 0.5321378707885742
1260
+ },
1261
+ {
1262
+ "name": "encoder.layers.12.self_attn.v_proj",
1263
+ "out_features": 1024,
1264
+ "in_features": 1024,
1265
+ "group_size": 128,
1266
+ "as_conv1d": false,
1267
+ "has_bias": false,
1268
+ "zero_fraction": 0.3993492126464844
1269
+ },
1270
+ {
1271
+ "name": "encoder.layers.12.self_attn.o_proj",
1272
+ "out_features": 1024,
1273
+ "in_features": 1024,
1274
+ "group_size": 128,
1275
+ "as_conv1d": false,
1276
+ "has_bias": false,
1277
+ "zero_fraction": 0.3653535842895508
1278
+ },
1279
+ {
1280
+ "name": "encoder.layers.12.self_attn.relative_k_proj",
1281
+ "out_features": 1024,
1282
+ "in_features": 1024,
1283
+ "group_size": 128,
1284
+ "as_conv1d": false,
1285
+ "has_bias": false,
1286
+ "zero_fraction": 0.16003036499023438
1287
+ },
1288
+ {
1289
+ "name": "encoder.layers.12.conv.pointwise_conv1",
1290
+ "out_features": 2048,
1291
+ "in_features": 1024,
1292
+ "group_size": 128,
1293
+ "as_conv1d": true,
1294
+ "has_bias": false,
1295
+ "zero_fraction": 0.43686389923095703
1296
+ },
1297
+ {
1298
+ "name": "encoder.layers.12.conv.pointwise_conv2",
1299
+ "out_features": 1024,
1300
+ "in_features": 1024,
1301
+ "group_size": 128,
1302
+ "as_conv1d": true,
1303
+ "has_bias": false,
1304
+ "zero_fraction": 0.4244871139526367
1305
+ },
1306
+ {
1307
+ "name": "encoder.layers.12.feed_forward2.linear1",
1308
+ "out_features": 4096,
1309
+ "in_features": 1024,
1310
+ "group_size": 128,
1311
+ "as_conv1d": false,
1312
+ "has_bias": false,
1313
+ "zero_fraction": 0.47611141204833984
1314
+ },
1315
+ {
1316
+ "name": "encoder.layers.12.feed_forward2.linear2",
1317
+ "out_features": 1024,
1318
+ "in_features": 4096,
1319
+ "group_size": 128,
1320
+ "as_conv1d": false,
1321
+ "has_bias": false,
1322
+ "zero_fraction": 0.5230574607849121
1323
+ },
1324
+ {
1325
+ "name": "encoder.layers.13.feed_forward1.linear1",
1326
+ "out_features": 4096,
1327
+ "in_features": 1024,
1328
+ "group_size": 128,
1329
+ "as_conv1d": false,
1330
+ "has_bias": false,
1331
+ "zero_fraction": 0.531564474105835
1332
+ },
1333
+ {
1334
+ "name": "encoder.layers.13.feed_forward1.linear2",
1335
+ "out_features": 1024,
1336
+ "in_features": 4096,
1337
+ "group_size": 128,
1338
+ "as_conv1d": false,
1339
+ "has_bias": false,
1340
+ "zero_fraction": 0.5122246742248535
1341
+ },
1342
+ {
1343
+ "name": "encoder.layers.13.self_attn.q_proj",
1344
+ "out_features": 1024,
1345
+ "in_features": 1024,
1346
+ "group_size": 128,
1347
+ "as_conv1d": false,
1348
+ "has_bias": false,
1349
+ "zero_fraction": 0.36064720153808594
1350
+ },
1351
+ {
1352
+ "name": "encoder.layers.13.self_attn.k_proj",
1353
+ "out_features": 1024,
1354
+ "in_features": 1024,
1355
+ "group_size": 128,
1356
+ "as_conv1d": false,
1357
+ "has_bias": false,
1358
+ "zero_fraction": 0.5534095764160156
1359
+ },
1360
+ {
1361
+ "name": "encoder.layers.13.self_attn.v_proj",
1362
+ "out_features": 1024,
1363
+ "in_features": 1024,
1364
+ "group_size": 128,
1365
+ "as_conv1d": false,
1366
+ "has_bias": false,
1367
+ "zero_fraction": 0.37798595428466797
1368
+ },
1369
+ {
1370
+ "name": "encoder.layers.13.self_attn.o_proj",
1371
+ "out_features": 1024,
1372
+ "in_features": 1024,
1373
+ "group_size": 128,
1374
+ "as_conv1d": false,
1375
+ "has_bias": false,
1376
+ "zero_fraction": 0.35416412353515625
1377
+ },
1378
+ {
1379
+ "name": "encoder.layers.13.self_attn.relative_k_proj",
1380
+ "out_features": 1024,
1381
+ "in_features": 1024,
1382
+ "group_size": 128,
1383
+ "as_conv1d": false,
1384
+ "has_bias": false,
1385
+ "zero_fraction": 0.14722442626953125
1386
+ },
1387
+ {
1388
+ "name": "encoder.layers.13.conv.pointwise_conv1",
1389
+ "out_features": 2048,
1390
+ "in_features": 1024,
1391
+ "group_size": 128,
1392
+ "as_conv1d": true,
1393
+ "has_bias": false,
1394
+ "zero_fraction": 0.43766117095947266
1395
+ },
1396
+ {
1397
+ "name": "encoder.layers.13.conv.pointwise_conv2",
1398
+ "out_features": 1024,
1399
+ "in_features": 1024,
1400
+ "group_size": 128,
1401
+ "as_conv1d": true,
1402
+ "has_bias": false,
1403
+ "zero_fraction": 0.41414833068847656
1404
+ },
1405
+ {
1406
+ "name": "encoder.layers.13.feed_forward2.linear1",
1407
+ "out_features": 4096,
1408
+ "in_features": 1024,
1409
+ "group_size": 128,
1410
+ "as_conv1d": false,
1411
+ "has_bias": false,
1412
+ "zero_fraction": 0.4815671443939209
1413
+ },
1414
+ {
1415
+ "name": "encoder.layers.13.feed_forward2.linear2",
1416
+ "out_features": 1024,
1417
+ "in_features": 4096,
1418
+ "group_size": 128,
1419
+ "as_conv1d": false,
1420
+ "has_bias": false,
1421
+ "zero_fraction": 0.4973268508911133
1422
+ },
1423
+ {
1424
+ "name": "encoder.layers.14.feed_forward1.linear1",
1425
+ "out_features": 4096,
1426
+ "in_features": 1024,
1427
+ "group_size": 128,
1428
+ "as_conv1d": false,
1429
+ "has_bias": false,
1430
+ "zero_fraction": 0.5406713485717773
1431
+ },
1432
+ {
1433
+ "name": "encoder.layers.14.feed_forward1.linear2",
1434
+ "out_features": 1024,
1435
+ "in_features": 4096,
1436
+ "group_size": 128,
1437
+ "as_conv1d": false,
1438
+ "has_bias": false,
1439
+ "zero_fraction": 0.520122766494751
1440
+ },
1441
+ {
1442
+ "name": "encoder.layers.14.self_attn.q_proj",
1443
+ "out_features": 1024,
1444
+ "in_features": 1024,
1445
+ "group_size": 128,
1446
+ "as_conv1d": false,
1447
+ "has_bias": false,
1448
+ "zero_fraction": 0.37984371185302734
1449
+ },
1450
+ {
1451
+ "name": "encoder.layers.14.self_attn.k_proj",
1452
+ "out_features": 1024,
1453
+ "in_features": 1024,
1454
+ "group_size": 128,
1455
+ "as_conv1d": false,
1456
+ "has_bias": false,
1457
+ "zero_fraction": 0.5693721771240234
1458
+ },
1459
+ {
1460
+ "name": "encoder.layers.14.self_attn.v_proj",
1461
+ "out_features": 1024,
1462
+ "in_features": 1024,
1463
+ "group_size": 128,
1464
+ "as_conv1d": false,
1465
+ "has_bias": false,
1466
+ "zero_fraction": 0.37279415130615234
1467
+ },
1468
+ {
1469
+ "name": "encoder.layers.14.self_attn.o_proj",
1470
+ "out_features": 1024,
1471
+ "in_features": 1024,
1472
+ "group_size": 128,
1473
+ "as_conv1d": false,
1474
+ "has_bias": false,
1475
+ "zero_fraction": 0.35245513916015625
1476
+ },
1477
+ {
1478
+ "name": "encoder.layers.14.self_attn.relative_k_proj",
1479
+ "out_features": 1024,
1480
+ "in_features": 1024,
1481
+ "group_size": 128,
1482
+ "as_conv1d": false,
1483
+ "has_bias": false,
1484
+ "zero_fraction": 0.14053821563720703
1485
+ },
1486
+ {
1487
+ "name": "encoder.layers.14.conv.pointwise_conv1",
1488
+ "out_features": 2048,
1489
+ "in_features": 1024,
1490
+ "group_size": 128,
1491
+ "as_conv1d": true,
1492
+ "has_bias": false,
1493
+ "zero_fraction": 0.4381241798400879
1494
+ },
1495
+ {
1496
+ "name": "encoder.layers.14.conv.pointwise_conv2",
1497
+ "out_features": 1024,
1498
+ "in_features": 1024,
1499
+ "group_size": 128,
1500
+ "as_conv1d": true,
1501
+ "has_bias": false,
1502
+ "zero_fraction": 0.41631603240966797
1503
+ },
1504
+ {
1505
+ "name": "encoder.layers.14.feed_forward2.linear1",
1506
+ "out_features": 4096,
1507
+ "in_features": 1024,
1508
+ "group_size": 128,
1509
+ "as_conv1d": false,
1510
+ "has_bias": false,
1511
+ "zero_fraction": 0.48996448516845703
1512
+ },
1513
+ {
1514
+ "name": "encoder.layers.14.feed_forward2.linear2",
1515
+ "out_features": 1024,
1516
+ "in_features": 4096,
1517
+ "group_size": 128,
1518
+ "as_conv1d": false,
1519
+ "has_bias": false,
1520
+ "zero_fraction": 0.5002038478851318
1521
+ },
1522
+ {
1523
+ "name": "encoder.layers.15.feed_forward1.linear1",
1524
+ "out_features": 4096,
1525
+ "in_features": 1024,
1526
+ "group_size": 128,
1527
+ "as_conv1d": false,
1528
+ "has_bias": false,
1529
+ "zero_fraction": 0.5357084274291992
1530
+ },
1531
+ {
1532
+ "name": "encoder.layers.15.feed_forward1.linear2",
1533
+ "out_features": 1024,
1534
+ "in_features": 4096,
1535
+ "group_size": 128,
1536
+ "as_conv1d": false,
1537
+ "has_bias": false,
1538
+ "zero_fraction": 0.5198726654052734
1539
+ },
1540
+ {
1541
+ "name": "encoder.layers.15.self_attn.q_proj",
1542
+ "out_features": 1024,
1543
+ "in_features": 1024,
1544
+ "group_size": 128,
1545
+ "as_conv1d": false,
1546
+ "has_bias": false,
1547
+ "zero_fraction": 0.3787193298339844
1548
+ },
1549
+ {
1550
+ "name": "encoder.layers.15.self_attn.k_proj",
1551
+ "out_features": 1024,
1552
+ "in_features": 1024,
1553
+ "group_size": 128,
1554
+ "as_conv1d": false,
1555
+ "has_bias": false,
1556
+ "zero_fraction": 0.6187000274658203
1557
+ },
1558
+ {
1559
+ "name": "encoder.layers.15.self_attn.v_proj",
1560
+ "out_features": 1024,
1561
+ "in_features": 1024,
1562
+ "group_size": 128,
1563
+ "as_conv1d": false,
1564
+ "has_bias": false,
1565
+ "zero_fraction": 0.37151432037353516
1566
+ },
1567
+ {
1568
+ "name": "encoder.layers.15.self_attn.o_proj",
1569
+ "out_features": 1024,
1570
+ "in_features": 1024,
1571
+ "group_size": 128,
1572
+ "as_conv1d": false,
1573
+ "has_bias": false,
1574
+ "zero_fraction": 0.3600912094116211
1575
+ },
1576
+ {
1577
+ "name": "encoder.layers.15.self_attn.relative_k_proj",
1578
+ "out_features": 1024,
1579
+ "in_features": 1024,
1580
+ "group_size": 128,
1581
+ "as_conv1d": false,
1582
+ "has_bias": false,
1583
+ "zero_fraction": 0.14568805694580078
1584
+ },
1585
+ {
1586
+ "name": "encoder.layers.15.conv.pointwise_conv1",
1587
+ "out_features": 2048,
1588
+ "in_features": 1024,
1589
+ "group_size": 128,
1590
+ "as_conv1d": true,
1591
+ "has_bias": false,
1592
+ "zero_fraction": 0.43272924423217773
1593
+ },
1594
+ {
1595
+ "name": "encoder.layers.15.conv.pointwise_conv2",
1596
+ "out_features": 1024,
1597
+ "in_features": 1024,
1598
+ "group_size": 128,
1599
+ "as_conv1d": true,
1600
+ "has_bias": false,
1601
+ "zero_fraction": 0.41810035705566406
1602
+ },
1603
+ {
1604
+ "name": "encoder.layers.15.feed_forward2.linear1",
1605
+ "out_features": 4096,
1606
+ "in_features": 1024,
1607
+ "group_size": 128,
1608
+ "as_conv1d": false,
1609
+ "has_bias": false,
1610
+ "zero_fraction": 0.49303531646728516
1611
+ },
1612
+ {
1613
+ "name": "encoder.layers.15.feed_forward2.linear2",
1614
+ "out_features": 1024,
1615
+ "in_features": 4096,
1616
+ "group_size": 128,
1617
+ "as_conv1d": false,
1618
+ "has_bias": false,
1619
+ "zero_fraction": 0.53395676612854
1620
+ },
1621
+ {
1622
+ "name": "encoder.layers.16.feed_forward1.linear1",
1623
+ "out_features": 4096,
1624
+ "in_features": 1024,
1625
+ "group_size": 128,
1626
+ "as_conv1d": false,
1627
+ "has_bias": false,
1628
+ "zero_fraction": 0.5362865924835205
1629
+ },
1630
+ {
1631
+ "name": "encoder.layers.16.feed_forward1.linear2",
1632
+ "out_features": 1024,
1633
+ "in_features": 4096,
1634
+ "group_size": 128,
1635
+ "as_conv1d": false,
1636
+ "has_bias": false,
1637
+ "zero_fraction": 0.5048935413360596
1638
+ },
1639
+ {
1640
+ "name": "encoder.layers.16.self_attn.q_proj",
1641
+ "out_features": 1024,
1642
+ "in_features": 1024,
1643
+ "group_size": 128,
1644
+ "as_conv1d": false,
1645
+ "has_bias": false,
1646
+ "zero_fraction": 0.3586082458496094
1647
+ },
1648
+ {
1649
+ "name": "encoder.layers.16.self_attn.k_proj",
1650
+ "out_features": 1024,
1651
+ "in_features": 1024,
1652
+ "group_size": 128,
1653
+ "as_conv1d": false,
1654
+ "has_bias": false,
1655
+ "zero_fraction": 0.5664215087890625
1656
+ },
1657
+ {
1658
+ "name": "encoder.layers.16.self_attn.v_proj",
1659
+ "out_features": 1024,
1660
+ "in_features": 1024,
1661
+ "group_size": 128,
1662
+ "as_conv1d": false,
1663
+ "has_bias": false,
1664
+ "zero_fraction": 0.3739347457885742
1665
+ },
1666
+ {
1667
+ "name": "encoder.layers.16.self_attn.o_proj",
1668
+ "out_features": 1024,
1669
+ "in_features": 1024,
1670
+ "group_size": 128,
1671
+ "as_conv1d": false,
1672
+ "has_bias": false,
1673
+ "zero_fraction": 0.36450672149658203
1674
+ },
1675
+ {
1676
+ "name": "encoder.layers.16.self_attn.relative_k_proj",
1677
+ "out_features": 1024,
1678
+ "in_features": 1024,
1679
+ "group_size": 128,
1680
+ "as_conv1d": false,
1681
+ "has_bias": false,
1682
+ "zero_fraction": 0.1494426727294922
1683
+ },
1684
+ {
1685
+ "name": "encoder.layers.16.conv.pointwise_conv1",
1686
+ "out_features": 2048,
1687
+ "in_features": 1024,
1688
+ "group_size": 128,
1689
+ "as_conv1d": true,
1690
+ "has_bias": false,
1691
+ "zero_fraction": 0.43798398971557617
1692
+ },
1693
+ {
1694
+ "name": "encoder.layers.16.conv.pointwise_conv2",
1695
+ "out_features": 1024,
1696
+ "in_features": 1024,
1697
+ "group_size": 128,
1698
+ "as_conv1d": true,
1699
+ "has_bias": false,
1700
+ "zero_fraction": 0.3995380401611328
1701
+ },
1702
+ {
1703
+ "name": "encoder.layers.16.feed_forward2.linear1",
1704
+ "out_features": 4096,
1705
+ "in_features": 1024,
1706
+ "group_size": 128,
1707
+ "as_conv1d": false,
1708
+ "has_bias": false,
1709
+ "zero_fraction": 0.46408939361572266
1710
+ },
1711
+ {
1712
+ "name": "encoder.layers.16.feed_forward2.linear2",
1713
+ "out_features": 1024,
1714
+ "in_features": 4096,
1715
+ "group_size": 128,
1716
+ "as_conv1d": false,
1717
+ "has_bias": false,
1718
+ "zero_fraction": 0.47188329696655273
1719
+ },
1720
+ {
1721
+ "name": "encoder.layers.17.feed_forward1.linear1",
1722
+ "out_features": 4096,
1723
+ "in_features": 1024,
1724
+ "group_size": 128,
1725
+ "as_conv1d": false,
1726
+ "has_bias": false,
1727
+ "zero_fraction": 0.4892551898956299
1728
+ },
1729
+ {
1730
+ "name": "encoder.layers.17.feed_forward1.linear2",
1731
+ "out_features": 1024,
1732
+ "in_features": 4096,
1733
+ "group_size": 128,
1734
+ "as_conv1d": false,
1735
+ "has_bias": false,
1736
+ "zero_fraction": 0.4277915954589844
1737
+ },
1738
+ {
1739
+ "name": "encoder.layers.17.self_attn.q_proj",
1740
+ "out_features": 1024,
1741
+ "in_features": 1024,
1742
+ "group_size": 128,
1743
+ "as_conv1d": false,
1744
+ "has_bias": false,
1745
+ "zero_fraction": 0.35779476165771484
1746
+ },
1747
+ {
1748
+ "name": "encoder.layers.17.self_attn.k_proj",
1749
+ "out_features": 1024,
1750
+ "in_features": 1024,
1751
+ "group_size": 128,
1752
+ "as_conv1d": false,
1753
+ "has_bias": false,
1754
+ "zero_fraction": 0.6346902847290039
1755
+ },
1756
+ {
1757
+ "name": "encoder.layers.17.self_attn.v_proj",
1758
+ "out_features": 1024,
1759
+ "in_features": 1024,
1760
+ "group_size": 128,
1761
+ "as_conv1d": false,
1762
+ "has_bias": false,
1763
+ "zero_fraction": 0.3553304672241211
1764
+ },
1765
+ {
1766
+ "name": "encoder.layers.17.self_attn.o_proj",
1767
+ "out_features": 1024,
1768
+ "in_features": 1024,
1769
+ "group_size": 128,
1770
+ "as_conv1d": false,
1771
+ "has_bias": false,
1772
+ "zero_fraction": 0.3436117172241211
1773
+ },
1774
+ {
1775
+ "name": "encoder.layers.17.self_attn.relative_k_proj",
1776
+ "out_features": 1024,
1777
+ "in_features": 1024,
1778
+ "group_size": 128,
1779
+ "as_conv1d": false,
1780
+ "has_bias": false,
1781
+ "zero_fraction": 0.2155017852783203
1782
+ },
1783
+ {
1784
+ "name": "encoder.layers.17.conv.pointwise_conv1",
1785
+ "out_features": 2048,
1786
+ "in_features": 1024,
1787
+ "group_size": 128,
1788
+ "as_conv1d": true,
1789
+ "has_bias": false,
1790
+ "zero_fraction": 0.4304986000061035
1791
+ },
1792
+ {
1793
+ "name": "encoder.layers.17.conv.pointwise_conv2",
1794
+ "out_features": 1024,
1795
+ "in_features": 1024,
1796
+ "group_size": 128,
1797
+ "as_conv1d": true,
1798
+ "has_bias": false,
1799
+ "zero_fraction": 0.39414215087890625
1800
+ },
1801
+ {
1802
+ "name": "encoder.layers.17.feed_forward2.linear1",
1803
+ "out_features": 4096,
1804
+ "in_features": 1024,
1805
+ "group_size": 128,
1806
+ "as_conv1d": false,
1807
+ "has_bias": false,
1808
+ "zero_fraction": 0.4398188591003418
1809
+ },
1810
+ {
1811
+ "name": "encoder.layers.17.feed_forward2.linear2",
1812
+ "out_features": 1024,
1813
+ "in_features": 4096,
1814
+ "group_size": 128,
1815
+ "as_conv1d": false,
1816
+ "has_bias": false,
1817
+ "zero_fraction": 0.4505887031555176
1818
+ },
1819
+ {
1820
+ "name": "encoder.layers.18.feed_forward1.linear1",
1821
+ "out_features": 4096,
1822
+ "in_features": 1024,
1823
+ "group_size": 128,
1824
+ "as_conv1d": false,
1825
+ "has_bias": false,
1826
+ "zero_fraction": 0.4843440055847168
1827
+ },
1828
+ {
1829
+ "name": "encoder.layers.18.feed_forward1.linear2",
1830
+ "out_features": 1024,
1831
+ "in_features": 4096,
1832
+ "group_size": 128,
1833
+ "as_conv1d": false,
1834
+ "has_bias": false,
1835
+ "zero_fraction": 0.43118953704833984
1836
+ },
1837
+ {
1838
+ "name": "encoder.layers.18.self_attn.q_proj",
1839
+ "out_features": 1024,
1840
+ "in_features": 1024,
1841
+ "group_size": 128,
1842
+ "as_conv1d": false,
1843
+ "has_bias": false,
1844
+ "zero_fraction": 0.3386106491088867
1845
+ },
1846
+ {
1847
+ "name": "encoder.layers.18.self_attn.k_proj",
1848
+ "out_features": 1024,
1849
+ "in_features": 1024,
1850
+ "group_size": 128,
1851
+ "as_conv1d": false,
1852
+ "has_bias": false,
1853
+ "zero_fraction": 0.5647468566894531
1854
+ },
1855
+ {
1856
+ "name": "encoder.layers.18.self_attn.v_proj",
1857
+ "out_features": 1024,
1858
+ "in_features": 1024,
1859
+ "group_size": 128,
1860
+ "as_conv1d": false,
1861
+ "has_bias": false,
1862
+ "zero_fraction": 0.37237071990966797
1863
+ },
1864
+ {
1865
+ "name": "encoder.layers.18.self_attn.o_proj",
1866
+ "out_features": 1024,
1867
+ "in_features": 1024,
1868
+ "group_size": 128,
1869
+ "as_conv1d": false,
1870
+ "has_bias": false,
1871
+ "zero_fraction": 0.3634309768676758
1872
+ },
1873
+ {
1874
+ "name": "encoder.layers.18.self_attn.relative_k_proj",
1875
+ "out_features": 1024,
1876
+ "in_features": 1024,
1877
+ "group_size": 128,
1878
+ "as_conv1d": false,
1879
+ "has_bias": false,
1880
+ "zero_fraction": 0.1491680145263672
1881
+ },
1882
+ {
1883
+ "name": "encoder.layers.18.conv.pointwise_conv1",
1884
+ "out_features": 2048,
1885
+ "in_features": 1024,
1886
+ "group_size": 128,
1887
+ "as_conv1d": true,
1888
+ "has_bias": false,
1889
+ "zero_fraction": 0.43354082107543945
1890
+ },
1891
+ {
1892
+ "name": "encoder.layers.18.conv.pointwise_conv2",
1893
+ "out_features": 1024,
1894
+ "in_features": 1024,
1895
+ "group_size": 128,
1896
+ "as_conv1d": true,
1897
+ "has_bias": false,
1898
+ "zero_fraction": 0.4006175994873047
1899
+ },
1900
+ {
1901
+ "name": "encoder.layers.18.feed_forward2.linear1",
1902
+ "out_features": 4096,
1903
+ "in_features": 1024,
1904
+ "group_size": 128,
1905
+ "as_conv1d": false,
1906
+ "has_bias": false,
1907
+ "zero_fraction": 0.4384031295776367
1908
+ },
1909
+ {
1910
+ "name": "encoder.layers.18.feed_forward2.linear2",
1911
+ "out_features": 1024,
1912
+ "in_features": 4096,
1913
+ "group_size": 128,
1914
+ "as_conv1d": false,
1915
+ "has_bias": false,
1916
+ "zero_fraction": 0.4480724334716797
1917
+ },
1918
+ {
1919
+ "name": "encoder.layers.19.feed_forward1.linear1",
1920
+ "out_features": 4096,
1921
+ "in_features": 1024,
1922
+ "group_size": 128,
1923
+ "as_conv1d": false,
1924
+ "has_bias": false,
1925
+ "zero_fraction": 0.47231221199035645
1926
+ },
1927
+ {
1928
+ "name": "encoder.layers.19.feed_forward1.linear2",
1929
+ "out_features": 1024,
1930
+ "in_features": 4096,
1931
+ "group_size": 128,
1932
+ "as_conv1d": false,
1933
+ "has_bias": false,
1934
+ "zero_fraction": 0.40505099296569824
1935
+ },
1936
+ {
1937
+ "name": "encoder.layers.19.self_attn.q_proj",
1938
+ "out_features": 1024,
1939
+ "in_features": 1024,
1940
+ "group_size": 128,
1941
+ "as_conv1d": false,
1942
+ "has_bias": false,
1943
+ "zero_fraction": 0.3529081344604492
1944
+ },
1945
+ {
1946
+ "name": "encoder.layers.19.self_attn.k_proj",
1947
+ "out_features": 1024,
1948
+ "in_features": 1024,
1949
+ "group_size": 128,
1950
+ "as_conv1d": false,
1951
+ "has_bias": false,
1952
+ "zero_fraction": 0.5809326171875
1953
+ },
1954
+ {
1955
+ "name": "encoder.layers.19.self_attn.v_proj",
1956
+ "out_features": 1024,
1957
+ "in_features": 1024,
1958
+ "group_size": 128,
1959
+ "as_conv1d": false,
1960
+ "has_bias": false,
1961
+ "zero_fraction": 0.36546993255615234
1962
+ },
1963
+ {
1964
+ "name": "encoder.layers.19.self_attn.o_proj",
1965
+ "out_features": 1024,
1966
+ "in_features": 1024,
1967
+ "group_size": 128,
1968
+ "as_conv1d": false,
1969
+ "has_bias": false,
1970
+ "zero_fraction": 0.35294342041015625
1971
+ },
1972
+ {
1973
+ "name": "encoder.layers.19.self_attn.relative_k_proj",
1974
+ "out_features": 1024,
1975
+ "in_features": 1024,
1976
+ "group_size": 128,
1977
+ "as_conv1d": false,
1978
+ "has_bias": false,
1979
+ "zero_fraction": 0.17095470428466797
1980
+ },
1981
+ {
1982
+ "name": "encoder.layers.19.conv.pointwise_conv1",
1983
+ "out_features": 2048,
1984
+ "in_features": 1024,
1985
+ "group_size": 128,
1986
+ "as_conv1d": true,
1987
+ "has_bias": false,
1988
+ "zero_fraction": 0.4216432571411133
1989
+ },
1990
+ {
1991
+ "name": "encoder.layers.19.conv.pointwise_conv2",
1992
+ "out_features": 1024,
1993
+ "in_features": 1024,
1994
+ "group_size": 128,
1995
+ "as_conv1d": true,
1996
+ "has_bias": false,
1997
+ "zero_fraction": 0.3793497085571289
1998
+ },
1999
+ {
2000
+ "name": "encoder.layers.19.feed_forward2.linear1",
2001
+ "out_features": 4096,
2002
+ "in_features": 1024,
2003
+ "group_size": 128,
2004
+ "as_conv1d": false,
2005
+ "has_bias": false,
2006
+ "zero_fraction": 0.42360615730285645
2007
+ },
2008
+ {
2009
+ "name": "encoder.layers.19.feed_forward2.linear2",
2010
+ "out_features": 1024,
2011
+ "in_features": 4096,
2012
+ "group_size": 128,
2013
+ "as_conv1d": false,
2014
+ "has_bias": false,
2015
+ "zero_fraction": 0.45482897758483887
2016
+ },
2017
+ {
2018
+ "name": "encoder.layers.20.feed_forward1.linear1",
2019
+ "out_features": 4096,
2020
+ "in_features": 1024,
2021
+ "group_size": 128,
2022
+ "as_conv1d": false,
2023
+ "has_bias": false,
2024
+ "zero_fraction": 0.4815986156463623
2025
+ },
2026
+ {
2027
+ "name": "encoder.layers.20.feed_forward1.linear2",
2028
+ "out_features": 1024,
2029
+ "in_features": 4096,
2030
+ "group_size": 128,
2031
+ "as_conv1d": false,
2032
+ "has_bias": false,
2033
+ "zero_fraction": 0.36296963691711426
2034
+ },
2035
+ {
2036
+ "name": "encoder.layers.20.self_attn.q_proj",
2037
+ "out_features": 1024,
2038
+ "in_features": 1024,
2039
+ "group_size": 128,
2040
+ "as_conv1d": false,
2041
+ "has_bias": false,
2042
+ "zero_fraction": 0.39092063903808594
2043
+ },
2044
+ {
2045
+ "name": "encoder.layers.20.self_attn.k_proj",
2046
+ "out_features": 1024,
2047
+ "in_features": 1024,
2048
+ "group_size": 128,
2049
+ "as_conv1d": false,
2050
+ "has_bias": false,
2051
+ "zero_fraction": 0.5931243896484375
2052
+ },
2053
+ {
2054
+ "name": "encoder.layers.20.self_attn.v_proj",
2055
+ "out_features": 1024,
2056
+ "in_features": 1024,
2057
+ "group_size": 128,
2058
+ "as_conv1d": false,
2059
+ "has_bias": false,
2060
+ "zero_fraction": 0.3127937316894531
2061
+ },
2062
+ {
2063
+ "name": "encoder.layers.20.self_attn.o_proj",
2064
+ "out_features": 1024,
2065
+ "in_features": 1024,
2066
+ "group_size": 128,
2067
+ "as_conv1d": false,
2068
+ "has_bias": false,
2069
+ "zero_fraction": 0.30979347229003906
2070
+ },
2071
+ {
2072
+ "name": "encoder.layers.20.self_attn.relative_k_proj",
2073
+ "out_features": 1024,
2074
+ "in_features": 1024,
2075
+ "group_size": 128,
2076
+ "as_conv1d": false,
2077
+ "has_bias": false,
2078
+ "zero_fraction": 0.14920997619628906
2079
+ },
2080
+ {
2081
+ "name": "encoder.layers.20.conv.pointwise_conv1",
2082
+ "out_features": 2048,
2083
+ "in_features": 1024,
2084
+ "group_size": 128,
2085
+ "as_conv1d": true,
2086
+ "has_bias": false,
2087
+ "zero_fraction": 0.4177713394165039
2088
+ },
2089
+ {
2090
+ "name": "encoder.layers.20.conv.pointwise_conv2",
2091
+ "out_features": 1024,
2092
+ "in_features": 1024,
2093
+ "group_size": 128,
2094
+ "as_conv1d": true,
2095
+ "has_bias": false,
2096
+ "zero_fraction": 0.33641719818115234
2097
+ },
2098
+ {
2099
+ "name": "encoder.layers.20.feed_forward2.linear1",
2100
+ "out_features": 4096,
2101
+ "in_features": 1024,
2102
+ "group_size": 128,
2103
+ "as_conv1d": false,
2104
+ "has_bias": false,
2105
+ "zero_fraction": 0.43540430068969727
2106
+ },
2107
+ {
2108
+ "name": "encoder.layers.20.feed_forward2.linear2",
2109
+ "out_features": 1024,
2110
+ "in_features": 4096,
2111
+ "group_size": 128,
2112
+ "as_conv1d": false,
2113
+ "has_bias": false,
2114
+ "zero_fraction": 0.39700770378112793
2115
+ },
2116
+ {
2117
+ "name": "encoder.layers.21.feed_forward1.linear1",
2118
+ "out_features": 4096,
2119
+ "in_features": 1024,
2120
+ "group_size": 128,
2121
+ "as_conv1d": false,
2122
+ "has_bias": false,
2123
+ "zero_fraction": 0.48596787452697754
2124
+ },
2125
+ {
2126
+ "name": "encoder.layers.21.feed_forward1.linear2",
2127
+ "out_features": 1024,
2128
+ "in_features": 4096,
2129
+ "group_size": 128,
2130
+ "as_conv1d": false,
2131
+ "has_bias": false,
2132
+ "zero_fraction": 0.3406181335449219
2133
+ },
2134
+ {
2135
+ "name": "encoder.layers.21.self_attn.q_proj",
2136
+ "out_features": 1024,
2137
+ "in_features": 1024,
2138
+ "group_size": 128,
2139
+ "as_conv1d": false,
2140
+ "has_bias": false,
2141
+ "zero_fraction": 0.37795257568359375
2142
+ },
2143
+ {
2144
+ "name": "encoder.layers.21.self_attn.k_proj",
2145
+ "out_features": 1024,
2146
+ "in_features": 1024,
2147
+ "group_size": 128,
2148
+ "as_conv1d": false,
2149
+ "has_bias": false,
2150
+ "zero_fraction": 0.5576667785644531
2151
+ },
2152
+ {
2153
+ "name": "encoder.layers.21.self_attn.v_proj",
2154
+ "out_features": 1024,
2155
+ "in_features": 1024,
2156
+ "group_size": 128,
2157
+ "as_conv1d": false,
2158
+ "has_bias": false,
2159
+ "zero_fraction": 0.3066892623901367
2160
+ },
2161
+ {
2162
+ "name": "encoder.layers.21.self_attn.o_proj",
2163
+ "out_features": 1024,
2164
+ "in_features": 1024,
2165
+ "group_size": 128,
2166
+ "as_conv1d": false,
2167
+ "has_bias": false,
2168
+ "zero_fraction": 0.2996206283569336
2169
+ },
2170
+ {
2171
+ "name": "encoder.layers.21.self_attn.relative_k_proj",
2172
+ "out_features": 1024,
2173
+ "in_features": 1024,
2174
+ "group_size": 128,
2175
+ "as_conv1d": false,
2176
+ "has_bias": false,
2177
+ "zero_fraction": 0.15436172485351562
2178
+ },
2179
+ {
2180
+ "name": "encoder.layers.21.conv.pointwise_conv1",
2181
+ "out_features": 2048,
2182
+ "in_features": 1024,
2183
+ "group_size": 128,
2184
+ "as_conv1d": true,
2185
+ "has_bias": false,
2186
+ "zero_fraction": 0.40650415420532227
2187
+ },
2188
+ {
2189
+ "name": "encoder.layers.21.conv.pointwise_conv2",
2190
+ "out_features": 1024,
2191
+ "in_features": 1024,
2192
+ "group_size": 128,
2193
+ "as_conv1d": true,
2194
+ "has_bias": false,
2195
+ "zero_fraction": 0.314422607421875
2196
+ },
2197
+ {
2198
+ "name": "encoder.layers.21.feed_forward2.linear1",
2199
+ "out_features": 4096,
2200
+ "in_features": 1024,
2201
+ "group_size": 128,
2202
+ "as_conv1d": false,
2203
+ "has_bias": false,
2204
+ "zero_fraction": 0.39052605628967285
2205
+ },
2206
+ {
2207
+ "name": "encoder.layers.21.feed_forward2.linear2",
2208
+ "out_features": 1024,
2209
+ "in_features": 4096,
2210
+ "group_size": 128,
2211
+ "as_conv1d": false,
2212
+ "has_bias": false,
2213
+ "zero_fraction": 0.3677656650543213
2214
+ },
2215
+ {
2216
+ "name": "encoder.layers.22.feed_forward1.linear1",
2217
+ "out_features": 4096,
2218
+ "in_features": 1024,
2219
+ "group_size": 128,
2220
+ "as_conv1d": false,
2221
+ "has_bias": false,
2222
+ "zero_fraction": 0.4427375793457031
2223
+ },
2224
+ {
2225
+ "name": "encoder.layers.22.feed_forward1.linear2",
2226
+ "out_features": 1024,
2227
+ "in_features": 4096,
2228
+ "group_size": 128,
2229
+ "as_conv1d": false,
2230
+ "has_bias": false,
2231
+ "zero_fraction": 0.3055906295776367
2232
+ },
2233
+ {
2234
+ "name": "encoder.layers.22.self_attn.q_proj",
2235
+ "out_features": 1024,
2236
+ "in_features": 1024,
2237
+ "group_size": 128,
2238
+ "as_conv1d": false,
2239
+ "has_bias": false,
2240
+ "zero_fraction": 0.3857412338256836
2241
+ },
2242
+ {
2243
+ "name": "encoder.layers.22.self_attn.k_proj",
2244
+ "out_features": 1024,
2245
+ "in_features": 1024,
2246
+ "group_size": 128,
2247
+ "as_conv1d": false,
2248
+ "has_bias": false,
2249
+ "zero_fraction": 0.5995121002197266
2250
+ },
2251
+ {
2252
+ "name": "encoder.layers.22.self_attn.v_proj",
2253
+ "out_features": 1024,
2254
+ "in_features": 1024,
2255
+ "group_size": 128,
2256
+ "as_conv1d": false,
2257
+ "has_bias": false,
2258
+ "zero_fraction": 0.2965202331542969
2259
+ },
2260
+ {
2261
+ "name": "encoder.layers.22.self_attn.o_proj",
2262
+ "out_features": 1024,
2263
+ "in_features": 1024,
2264
+ "group_size": 128,
2265
+ "as_conv1d": false,
2266
+ "has_bias": false,
2267
+ "zero_fraction": 0.28084754943847656
2268
+ },
2269
+ {
2270
+ "name": "encoder.layers.22.self_attn.relative_k_proj",
2271
+ "out_features": 1024,
2272
+ "in_features": 1024,
2273
+ "group_size": 128,
2274
+ "as_conv1d": false,
2275
+ "has_bias": false,
2276
+ "zero_fraction": 0.14216327667236328
2277
+ },
2278
+ {
2279
+ "name": "encoder.layers.22.conv.pointwise_conv1",
2280
+ "out_features": 2048,
2281
+ "in_features": 1024,
2282
+ "group_size": 128,
2283
+ "as_conv1d": true,
2284
+ "has_bias": false,
2285
+ "zero_fraction": 0.37828636169433594
2286
+ },
2287
+ {
2288
+ "name": "encoder.layers.22.conv.pointwise_conv2",
2289
+ "out_features": 1024,
2290
+ "in_features": 1024,
2291
+ "group_size": 128,
2292
+ "as_conv1d": true,
2293
+ "has_bias": false,
2294
+ "zero_fraction": 0.2952146530151367
2295
+ },
2296
+ {
2297
+ "name": "encoder.layers.22.feed_forward2.linear1",
2298
+ "out_features": 4096,
2299
+ "in_features": 1024,
2300
+ "group_size": 128,
2301
+ "as_conv1d": false,
2302
+ "has_bias": false,
2303
+ "zero_fraction": 0.4962489604949951
2304
+ },
2305
+ {
2306
+ "name": "encoder.layers.22.feed_forward2.linear2",
2307
+ "out_features": 1024,
2308
+ "in_features": 4096,
2309
+ "group_size": 128,
2310
+ "as_conv1d": false,
2311
+ "has_bias": false,
2312
+ "zero_fraction": 0.32378721237182617
2313
+ },
2314
+ {
2315
+ "name": "encoder.layers.23.feed_forward1.linear1",
2316
+ "out_features": 4096,
2317
+ "in_features": 1024,
2318
+ "group_size": 128,
2319
+ "as_conv1d": false,
2320
+ "has_bias": false,
2321
+ "zero_fraction": 0.4867126941680908
2322
+ },
2323
+ {
2324
+ "name": "encoder.layers.23.feed_forward1.linear2",
2325
+ "out_features": 1024,
2326
+ "in_features": 4096,
2327
+ "group_size": 128,
2328
+ "as_conv1d": false,
2329
+ "has_bias": false,
2330
+ "zero_fraction": 0.2900993824005127
2331
+ },
2332
+ {
2333
+ "name": "encoder.layers.23.self_attn.q_proj",
2334
+ "out_features": 1024,
2335
+ "in_features": 1024,
2336
+ "group_size": 128,
2337
+ "as_conv1d": false,
2338
+ "has_bias": false,
2339
+ "zero_fraction": 0.3742027282714844
2340
+ },
2341
+ {
2342
+ "name": "encoder.layers.23.self_attn.k_proj",
2343
+ "out_features": 1024,
2344
+ "in_features": 1024,
2345
+ "group_size": 128,
2346
+ "as_conv1d": false,
2347
+ "has_bias": false,
2348
+ "zero_fraction": 0.5684337615966797
2349
+ },
2350
+ {
2351
+ "name": "encoder.layers.23.self_attn.v_proj",
2352
+ "out_features": 1024,
2353
+ "in_features": 1024,
2354
+ "group_size": 128,
2355
+ "as_conv1d": false,
2356
+ "has_bias": false,
2357
+ "zero_fraction": 0.2869300842285156
2358
+ },
2359
+ {
2360
+ "name": "encoder.layers.23.self_attn.o_proj",
2361
+ "out_features": 1024,
2362
+ "in_features": 1024,
2363
+ "group_size": 128,
2364
+ "as_conv1d": false,
2365
+ "has_bias": false,
2366
+ "zero_fraction": 0.29657459259033203
2367
+ },
2368
+ {
2369
+ "name": "encoder.layers.23.self_attn.relative_k_proj",
2370
+ "out_features": 1024,
2371
+ "in_features": 1024,
2372
+ "group_size": 128,
2373
+ "as_conv1d": false,
2374
+ "has_bias": false,
2375
+ "zero_fraction": 0.1371021270751953
2376
+ },
2377
+ {
2378
+ "name": "encoder.layers.23.conv.pointwise_conv1",
2379
+ "out_features": 2048,
2380
+ "in_features": 1024,
2381
+ "group_size": 128,
2382
+ "as_conv1d": true,
2383
+ "has_bias": false,
2384
+ "zero_fraction": 0.38350915908813477
2385
+ },
2386
+ {
2387
+ "name": "encoder.layers.23.conv.pointwise_conv2",
2388
+ "out_features": 1024,
2389
+ "in_features": 1024,
2390
+ "group_size": 128,
2391
+ "as_conv1d": true,
2392
+ "has_bias": false,
2393
+ "zero_fraction": 0.3024635314941406
2394
+ },
2395
+ {
2396
+ "name": "encoder.layers.23.feed_forward2.linear1",
2397
+ "out_features": 4096,
2398
+ "in_features": 1024,
2399
+ "group_size": 128,
2400
+ "as_conv1d": false,
2401
+ "has_bias": false,
2402
+ "zero_fraction": 0.30291128158569336
2403
+ },
2404
+ {
2405
+ "name": "encoder.layers.23.feed_forward2.linear2",
2406
+ "out_features": 1024,
2407
+ "in_features": 4096,
2408
+ "group_size": 128,
2409
+ "as_conv1d": false,
2410
+ "has_bias": false,
2411
+ "zero_fraction": 0.3130035400390625
2412
+ }
2413
+ ],
2414
+ "n_quantized_params": 603979776,
2415
+ "n_dense_params": 23290911,
2416
+ "size_mb": 177.7
2417
+ }
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff