Upload folder using huggingface_hub
Browse files- .gitattributes +5 -0
- checkpoint-10000/chat_template.jinja +89 -0
- checkpoint-10000/config.json +101 -0
- checkpoint-10000/model.safetensors +3 -0
- checkpoint-10000/tokenizer.json +3 -0
- checkpoint-10000/tokenizer_config.json +23 -0
- checkpoint-10000/train_config.json +58 -0
- checkpoint-20000/chat_template.jinja +89 -0
- checkpoint-20000/config.json +101 -0
- checkpoint-20000/model.safetensors +3 -0
- checkpoint-20000/tokenizer.json +3 -0
- checkpoint-20000/tokenizer_config.json +23 -0
- checkpoint-20000/train_config.json +58 -0
- checkpoint-30000/chat_template.jinja +89 -0
- checkpoint-30000/config.json +101 -0
- checkpoint-30000/model.safetensors +3 -0
- checkpoint-30000/tokenizer.json +3 -0
- checkpoint-30000/tokenizer_config.json +23 -0
- checkpoint-30000/train_config.json +58 -0
- checkpoint-40000/chat_template.jinja +89 -0
- checkpoint-40000/config.json +101 -0
- checkpoint-40000/model.safetensors +3 -0
- checkpoint-40000/tokenizer.json +3 -0
- checkpoint-40000/tokenizer_config.json +23 -0
- checkpoint-40000/train_config.json +58 -0
- checkpoint-50000/chat_template.jinja +89 -0
- checkpoint-50000/config.json +101 -0
- checkpoint-50000/model.safetensors +3 -0
- checkpoint-50000/tokenizer.json +3 -0
- checkpoint-50000/tokenizer_config.json +23 -0
- checkpoint-50000/train_config.json +58 -0
- initial_config.json +58 -0
- tensorboard/events.out.tfevents.1777886498.osaka.2227663.0 +3 -0
- train.log +50 -0
.gitattributes
CHANGED
|
@@ -33,3 +33,8 @@ saved_model/**/* 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
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 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
|
| 36 |
+
checkpoint-10000/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 37 |
+
checkpoint-20000/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 38 |
+
checkpoint-30000/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 39 |
+
checkpoint-40000/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
| 40 |
+
checkpoint-50000/tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
checkpoint-10000/chat_template.jinja
ADDED
|
@@ -0,0 +1,89 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{%- if tools %}
|
| 2 |
+
{{- '<|im_start|>system\n' }}
|
| 3 |
+
{%- if messages[0].role == 'system' %}
|
| 4 |
+
{{- messages[0].content + '\n\n' }}
|
| 5 |
+
{%- endif %}
|
| 6 |
+
{{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
|
| 7 |
+
{%- for tool in tools %}
|
| 8 |
+
{{- "\n" }}
|
| 9 |
+
{{- tool | tojson }}
|
| 10 |
+
{%- endfor %}
|
| 11 |
+
{{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
|
| 12 |
+
{%- else %}
|
| 13 |
+
{%- if messages[0].role == 'system' %}
|
| 14 |
+
{{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
|
| 15 |
+
{%- endif %}
|
| 16 |
+
{%- endif %}
|
| 17 |
+
{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
|
| 18 |
+
{%- for message in messages[::-1] %}
|
| 19 |
+
{%- set index = (messages|length - 1) - loop.index0 %}
|
| 20 |
+
{%- if ns.multi_step_tool and message.role == "user" and message.content is string and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
|
| 21 |
+
{%- set ns.multi_step_tool = false %}
|
| 22 |
+
{%- set ns.last_query_index = index %}
|
| 23 |
+
{%- endif %}
|
| 24 |
+
{%- endfor %}
|
| 25 |
+
{%- for message in messages %}
|
| 26 |
+
{%- if message.content is string %}
|
| 27 |
+
{%- set content = message.content %}
|
| 28 |
+
{%- else %}
|
| 29 |
+
{%- set content = '' %}
|
| 30 |
+
{%- endif %}
|
| 31 |
+
{%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
|
| 32 |
+
{{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
|
| 33 |
+
{%- elif message.role == "assistant" %}
|
| 34 |
+
{%- set reasoning_content = '' %}
|
| 35 |
+
{%- if message.reasoning_content is string %}
|
| 36 |
+
{%- set reasoning_content = message.reasoning_content %}
|
| 37 |
+
{%- else %}
|
| 38 |
+
{%- if '</think>' in content %}
|
| 39 |
+
{%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
|
| 40 |
+
{%- set content = content.split('</think>')[-1].lstrip('\n') %}
|
| 41 |
+
{%- endif %}
|
| 42 |
+
{%- endif %}
|
| 43 |
+
{%- if loop.index0 > ns.last_query_index %}
|
| 44 |
+
{%- if loop.last or (not loop.last and reasoning_content) %}
|
| 45 |
+
{{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
|
| 46 |
+
{%- else %}
|
| 47 |
+
{{- '<|im_start|>' + message.role + '\n' + content }}
|
| 48 |
+
{%- endif %}
|
| 49 |
+
{%- else %}
|
| 50 |
+
{{- '<|im_start|>' + message.role + '\n' + content }}
|
| 51 |
+
{%- endif %}
|
| 52 |
+
{%- if message.tool_calls %}
|
| 53 |
+
{%- for tool_call in message.tool_calls %}
|
| 54 |
+
{%- if (loop.first and content) or (not loop.first) %}
|
| 55 |
+
{{- '\n' }}
|
| 56 |
+
{%- endif %}
|
| 57 |
+
{%- if tool_call.function %}
|
| 58 |
+
{%- set tool_call = tool_call.function %}
|
| 59 |
+
{%- endif %}
|
| 60 |
+
{{- '<tool_call>\n{"name": "' }}
|
| 61 |
+
{{- tool_call.name }}
|
| 62 |
+
{{- '", "arguments": ' }}
|
| 63 |
+
{%- if tool_call.arguments is string %}
|
| 64 |
+
{{- tool_call.arguments }}
|
| 65 |
+
{%- else %}
|
| 66 |
+
{{- tool_call.arguments | tojson }}
|
| 67 |
+
{%- endif %}
|
| 68 |
+
{{- '}\n</tool_call>' }}
|
| 69 |
+
{%- endfor %}
|
| 70 |
+
{%- endif %}
|
| 71 |
+
{{- '<|im_end|>\n' }}
|
| 72 |
+
{%- elif message.role == "tool" %}
|
| 73 |
+
{%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
|
| 74 |
+
{{- '<|im_start|>user' }}
|
| 75 |
+
{%- endif %}
|
| 76 |
+
{{- '\n<tool_response>\n' }}
|
| 77 |
+
{{- content }}
|
| 78 |
+
{{- '\n</tool_response>' }}
|
| 79 |
+
{%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
|
| 80 |
+
{{- '<|im_end|>\n' }}
|
| 81 |
+
{%- endif %}
|
| 82 |
+
{%- endif %}
|
| 83 |
+
{%- endfor %}
|
| 84 |
+
{%- if add_generation_prompt %}
|
| 85 |
+
{{- '<|im_start|>assistant\n' }}
|
| 86 |
+
{%- if enable_thinking is defined and enable_thinking is false %}
|
| 87 |
+
{{- '<think>\n\n</think>\n\n' }}
|
| 88 |
+
{%- endif %}
|
| 89 |
+
{%- endif %}
|
checkpoint-10000/config.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"OmniVoice"
|
| 4 |
+
],
|
| 5 |
+
"audio_codebook_weights": [
|
| 6 |
+
8,
|
| 7 |
+
8,
|
| 8 |
+
6,
|
| 9 |
+
6,
|
| 10 |
+
4,
|
| 11 |
+
4,
|
| 12 |
+
2,
|
| 13 |
+
2
|
| 14 |
+
],
|
| 15 |
+
"audio_mask_id": 1024,
|
| 16 |
+
"audio_vocab_size": 1025,
|
| 17 |
+
"bos_token_id": null,
|
| 18 |
+
"dtype": "float32",
|
| 19 |
+
"eos_token_id": 151645,
|
| 20 |
+
"llm_config": {
|
| 21 |
+
"_name_or_path": "",
|
| 22 |
+
"architectures": [
|
| 23 |
+
"Qwen3ForCausalLM"
|
| 24 |
+
],
|
| 25 |
+
"attention_bias": false,
|
| 26 |
+
"attention_dropout": 0.0,
|
| 27 |
+
"bos_token_id": 151643,
|
| 28 |
+
"chunk_size_feed_forward": 0,
|
| 29 |
+
"dtype": "float32",
|
| 30 |
+
"eos_token_id": 151645,
|
| 31 |
+
"head_dim": 128,
|
| 32 |
+
"hidden_act": "silu",
|
| 33 |
+
"hidden_size": 1024,
|
| 34 |
+
"id2label": {
|
| 35 |
+
"0": "LABEL_0",
|
| 36 |
+
"1": "LABEL_1"
|
| 37 |
+
},
|
| 38 |
+
"initializer_range": 0.02,
|
| 39 |
+
"intermediate_size": 3072,
|
| 40 |
+
"is_encoder_decoder": false,
|
| 41 |
+
"label2id": {
|
| 42 |
+
"LABEL_0": 0,
|
| 43 |
+
"LABEL_1": 1
|
| 44 |
+
},
|
| 45 |
+
"layer_types": [
|
| 46 |
+
"full_attention",
|
| 47 |
+
"full_attention",
|
| 48 |
+
"full_attention",
|
| 49 |
+
"full_attention",
|
| 50 |
+
"full_attention",
|
| 51 |
+
"full_attention",
|
| 52 |
+
"full_attention",
|
| 53 |
+
"full_attention",
|
| 54 |
+
"full_attention",
|
| 55 |
+
"full_attention",
|
| 56 |
+
"full_attention",
|
| 57 |
+
"full_attention",
|
| 58 |
+
"full_attention",
|
| 59 |
+
"full_attention",
|
| 60 |
+
"full_attention",
|
| 61 |
+
"full_attention",
|
| 62 |
+
"full_attention",
|
| 63 |
+
"full_attention",
|
| 64 |
+
"full_attention",
|
| 65 |
+
"full_attention",
|
| 66 |
+
"full_attention",
|
| 67 |
+
"full_attention",
|
| 68 |
+
"full_attention",
|
| 69 |
+
"full_attention",
|
| 70 |
+
"full_attention",
|
| 71 |
+
"full_attention",
|
| 72 |
+
"full_attention",
|
| 73 |
+
"full_attention"
|
| 74 |
+
],
|
| 75 |
+
"max_position_embeddings": 40960,
|
| 76 |
+
"max_window_layers": 28,
|
| 77 |
+
"model_type": "qwen3",
|
| 78 |
+
"num_attention_heads": 16,
|
| 79 |
+
"num_hidden_layers": 28,
|
| 80 |
+
"num_key_value_heads": 8,
|
| 81 |
+
"output_attentions": false,
|
| 82 |
+
"output_hidden_states": false,
|
| 83 |
+
"pad_token_id": null,
|
| 84 |
+
"problem_type": null,
|
| 85 |
+
"return_dict": true,
|
| 86 |
+
"rms_norm_eps": 1e-06,
|
| 87 |
+
"rope_parameters": {
|
| 88 |
+
"rope_theta": 1000000,
|
| 89 |
+
"rope_type": "default"
|
| 90 |
+
},
|
| 91 |
+
"sliding_window": null,
|
| 92 |
+
"tie_word_embeddings": true,
|
| 93 |
+
"use_cache": true,
|
| 94 |
+
"use_sliding_window": false,
|
| 95 |
+
"vocab_size": 151676
|
| 96 |
+
},
|
| 97 |
+
"model_type": "omnivoice",
|
| 98 |
+
"num_audio_codebook": 8,
|
| 99 |
+
"pad_token_id": 151643,
|
| 100 |
+
"transformers_version": "5.3.0"
|
| 101 |
+
}
|
checkpoint-10000/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:892601425de2e97ca82b30c786319747bd983e12d262d5df0a3853fc32ac5e99
|
| 3 |
+
size 2450344144
|
checkpoint-10000/tokenizer.json
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:408f669b7e2b045fdf54201d815bd364e6667dbd845115da81239c40bc6dcfd1
|
| 3 |
+
size 11423986
|
checkpoint-10000/tokenizer_config.json
ADDED
|
@@ -0,0 +1,23 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": null,
|
| 5 |
+
"clean_up_tokenization_spaces": false,
|
| 6 |
+
"eos_token": "<|im_end|>",
|
| 7 |
+
"errors": "replace",
|
| 8 |
+
"extra_special_tokens": [
|
| 9 |
+
"<|denoise|>",
|
| 10 |
+
"<|lang_start|>",
|
| 11 |
+
"<|lang_end|>",
|
| 12 |
+
"<|instruct_start|>",
|
| 13 |
+
"<|instruct_end|>",
|
| 14 |
+
"<|text_start|>",
|
| 15 |
+
"<|text_end|>"
|
| 16 |
+
],
|
| 17 |
+
"is_local": true,
|
| 18 |
+
"model_max_length": 131072,
|
| 19 |
+
"pad_token": "<|endoftext|>",
|
| 20 |
+
"split_special_tokens": false,
|
| 21 |
+
"tokenizer_class": "Qwen2Tokenizer",
|
| 22 |
+
"unk_token": null
|
| 23 |
+
}
|
checkpoint-10000/train_config.json
ADDED
|
@@ -0,0 +1,58 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"output_dir": "exp/omnivoice_vietnamese_4kh",
|
| 3 |
+
"data_config": "/vast/tts/robert/OmniVoice/data/data_config_vietnamese.json",
|
| 4 |
+
"llm_name_or_path": "Qwen/Qwen3-0.6B",
|
| 5 |
+
"audio_vocab_size": 1025,
|
| 6 |
+
"audio_mask_id": 1024,
|
| 7 |
+
"num_audio_codebook": 8,
|
| 8 |
+
"audio_codebook_weights": [
|
| 9 |
+
8,
|
| 10 |
+
8,
|
| 11 |
+
6,
|
| 12 |
+
6,
|
| 13 |
+
4,
|
| 14 |
+
4,
|
| 15 |
+
2,
|
| 16 |
+
2
|
| 17 |
+
],
|
| 18 |
+
"drop_cond_ratio": 0.1,
|
| 19 |
+
"prompt_ratio_range": [
|
| 20 |
+
0.0,
|
| 21 |
+
0.3
|
| 22 |
+
],
|
| 23 |
+
"mask_ratio_range": [
|
| 24 |
+
0.0,
|
| 25 |
+
1.0
|
| 26 |
+
],
|
| 27 |
+
"language_ratio": 0.8,
|
| 28 |
+
"use_pinyin_ratio": 0.0,
|
| 29 |
+
"instruct_ratio": 0.0,
|
| 30 |
+
"only_instruct_ratio": 0.0,
|
| 31 |
+
"resume_from_checkpoint": null,
|
| 32 |
+
"init_from_checkpoint": "k2-fsa/OmniVoice",
|
| 33 |
+
"learning_rate": 1e-05,
|
| 34 |
+
"weight_decay": 0.01,
|
| 35 |
+
"max_grad_norm": 1.0,
|
| 36 |
+
"steps": 100000,
|
| 37 |
+
"epochs": null,
|
| 38 |
+
"seed": 42,
|
| 39 |
+
"lr_scheduler_type": "cosine",
|
| 40 |
+
"warmup_type": "ratio",
|
| 41 |
+
"warmup_ratio": 0.01,
|
| 42 |
+
"warmup_steps": 0,
|
| 43 |
+
"batch_tokens": 8192,
|
| 44 |
+
"gradient_accumulation_steps": 1,
|
| 45 |
+
"num_workers": 2,
|
| 46 |
+
"mixed_precision": "bf16",
|
| 47 |
+
"allow_tf32": true,
|
| 48 |
+
"use_deepspeed": false,
|
| 49 |
+
"deepspeed_config": null,
|
| 50 |
+
"attn_implementation": "flex_attention",
|
| 51 |
+
"max_sample_tokens": 2000,
|
| 52 |
+
"min_sample_tokens": 50,
|
| 53 |
+
"max_batch_size": 64,
|
| 54 |
+
"logging_steps": 50,
|
| 55 |
+
"eval_steps": 10000,
|
| 56 |
+
"save_steps": 10000,
|
| 57 |
+
"keep_last_n_checkpoints": -1
|
| 58 |
+
}
|
checkpoint-20000/chat_template.jinja
ADDED
|
@@ -0,0 +1,89 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{%- if tools %}
|
| 2 |
+
{{- '<|im_start|>system\n' }}
|
| 3 |
+
{%- if messages[0].role == 'system' %}
|
| 4 |
+
{{- messages[0].content + '\n\n' }}
|
| 5 |
+
{%- endif %}
|
| 6 |
+
{{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
|
| 7 |
+
{%- for tool in tools %}
|
| 8 |
+
{{- "\n" }}
|
| 9 |
+
{{- tool | tojson }}
|
| 10 |
+
{%- endfor %}
|
| 11 |
+
{{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
|
| 12 |
+
{%- else %}
|
| 13 |
+
{%- if messages[0].role == 'system' %}
|
| 14 |
+
{{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
|
| 15 |
+
{%- endif %}
|
| 16 |
+
{%- endif %}
|
| 17 |
+
{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
|
| 18 |
+
{%- for message in messages[::-1] %}
|
| 19 |
+
{%- set index = (messages|length - 1) - loop.index0 %}
|
| 20 |
+
{%- if ns.multi_step_tool and message.role == "user" and message.content is string and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
|
| 21 |
+
{%- set ns.multi_step_tool = false %}
|
| 22 |
+
{%- set ns.last_query_index = index %}
|
| 23 |
+
{%- endif %}
|
| 24 |
+
{%- endfor %}
|
| 25 |
+
{%- for message in messages %}
|
| 26 |
+
{%- if message.content is string %}
|
| 27 |
+
{%- set content = message.content %}
|
| 28 |
+
{%- else %}
|
| 29 |
+
{%- set content = '' %}
|
| 30 |
+
{%- endif %}
|
| 31 |
+
{%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
|
| 32 |
+
{{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
|
| 33 |
+
{%- elif message.role == "assistant" %}
|
| 34 |
+
{%- set reasoning_content = '' %}
|
| 35 |
+
{%- if message.reasoning_content is string %}
|
| 36 |
+
{%- set reasoning_content = message.reasoning_content %}
|
| 37 |
+
{%- else %}
|
| 38 |
+
{%- if '</think>' in content %}
|
| 39 |
+
{%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
|
| 40 |
+
{%- set content = content.split('</think>')[-1].lstrip('\n') %}
|
| 41 |
+
{%- endif %}
|
| 42 |
+
{%- endif %}
|
| 43 |
+
{%- if loop.index0 > ns.last_query_index %}
|
| 44 |
+
{%- if loop.last or (not loop.last and reasoning_content) %}
|
| 45 |
+
{{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
|
| 46 |
+
{%- else %}
|
| 47 |
+
{{- '<|im_start|>' + message.role + '\n' + content }}
|
| 48 |
+
{%- endif %}
|
| 49 |
+
{%- else %}
|
| 50 |
+
{{- '<|im_start|>' + message.role + '\n' + content }}
|
| 51 |
+
{%- endif %}
|
| 52 |
+
{%- if message.tool_calls %}
|
| 53 |
+
{%- for tool_call in message.tool_calls %}
|
| 54 |
+
{%- if (loop.first and content) or (not loop.first) %}
|
| 55 |
+
{{- '\n' }}
|
| 56 |
+
{%- endif %}
|
| 57 |
+
{%- if tool_call.function %}
|
| 58 |
+
{%- set tool_call = tool_call.function %}
|
| 59 |
+
{%- endif %}
|
| 60 |
+
{{- '<tool_call>\n{"name": "' }}
|
| 61 |
+
{{- tool_call.name }}
|
| 62 |
+
{{- '", "arguments": ' }}
|
| 63 |
+
{%- if tool_call.arguments is string %}
|
| 64 |
+
{{- tool_call.arguments }}
|
| 65 |
+
{%- else %}
|
| 66 |
+
{{- tool_call.arguments | tojson }}
|
| 67 |
+
{%- endif %}
|
| 68 |
+
{{- '}\n</tool_call>' }}
|
| 69 |
+
{%- endfor %}
|
| 70 |
+
{%- endif %}
|
| 71 |
+
{{- '<|im_end|>\n' }}
|
| 72 |
+
{%- elif message.role == "tool" %}
|
| 73 |
+
{%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
|
| 74 |
+
{{- '<|im_start|>user' }}
|
| 75 |
+
{%- endif %}
|
| 76 |
+
{{- '\n<tool_response>\n' }}
|
| 77 |
+
{{- content }}
|
| 78 |
+
{{- '\n</tool_response>' }}
|
| 79 |
+
{%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
|
| 80 |
+
{{- '<|im_end|>\n' }}
|
| 81 |
+
{%- endif %}
|
| 82 |
+
{%- endif %}
|
| 83 |
+
{%- endfor %}
|
| 84 |
+
{%- if add_generation_prompt %}
|
| 85 |
+
{{- '<|im_start|>assistant\n' }}
|
| 86 |
+
{%- if enable_thinking is defined and enable_thinking is false %}
|
| 87 |
+
{{- '<think>\n\n</think>\n\n' }}
|
| 88 |
+
{%- endif %}
|
| 89 |
+
{%- endif %}
|
checkpoint-20000/config.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"OmniVoice"
|
| 4 |
+
],
|
| 5 |
+
"audio_codebook_weights": [
|
| 6 |
+
8,
|
| 7 |
+
8,
|
| 8 |
+
6,
|
| 9 |
+
6,
|
| 10 |
+
4,
|
| 11 |
+
4,
|
| 12 |
+
2,
|
| 13 |
+
2
|
| 14 |
+
],
|
| 15 |
+
"audio_mask_id": 1024,
|
| 16 |
+
"audio_vocab_size": 1025,
|
| 17 |
+
"bos_token_id": null,
|
| 18 |
+
"dtype": "float32",
|
| 19 |
+
"eos_token_id": 151645,
|
| 20 |
+
"llm_config": {
|
| 21 |
+
"_name_or_path": "",
|
| 22 |
+
"architectures": [
|
| 23 |
+
"Qwen3ForCausalLM"
|
| 24 |
+
],
|
| 25 |
+
"attention_bias": false,
|
| 26 |
+
"attention_dropout": 0.0,
|
| 27 |
+
"bos_token_id": 151643,
|
| 28 |
+
"chunk_size_feed_forward": 0,
|
| 29 |
+
"dtype": "float32",
|
| 30 |
+
"eos_token_id": 151645,
|
| 31 |
+
"head_dim": 128,
|
| 32 |
+
"hidden_act": "silu",
|
| 33 |
+
"hidden_size": 1024,
|
| 34 |
+
"id2label": {
|
| 35 |
+
"0": "LABEL_0",
|
| 36 |
+
"1": "LABEL_1"
|
| 37 |
+
},
|
| 38 |
+
"initializer_range": 0.02,
|
| 39 |
+
"intermediate_size": 3072,
|
| 40 |
+
"is_encoder_decoder": false,
|
| 41 |
+
"label2id": {
|
| 42 |
+
"LABEL_0": 0,
|
| 43 |
+
"LABEL_1": 1
|
| 44 |
+
},
|
| 45 |
+
"layer_types": [
|
| 46 |
+
"full_attention",
|
| 47 |
+
"full_attention",
|
| 48 |
+
"full_attention",
|
| 49 |
+
"full_attention",
|
| 50 |
+
"full_attention",
|
| 51 |
+
"full_attention",
|
| 52 |
+
"full_attention",
|
| 53 |
+
"full_attention",
|
| 54 |
+
"full_attention",
|
| 55 |
+
"full_attention",
|
| 56 |
+
"full_attention",
|
| 57 |
+
"full_attention",
|
| 58 |
+
"full_attention",
|
| 59 |
+
"full_attention",
|
| 60 |
+
"full_attention",
|
| 61 |
+
"full_attention",
|
| 62 |
+
"full_attention",
|
| 63 |
+
"full_attention",
|
| 64 |
+
"full_attention",
|
| 65 |
+
"full_attention",
|
| 66 |
+
"full_attention",
|
| 67 |
+
"full_attention",
|
| 68 |
+
"full_attention",
|
| 69 |
+
"full_attention",
|
| 70 |
+
"full_attention",
|
| 71 |
+
"full_attention",
|
| 72 |
+
"full_attention",
|
| 73 |
+
"full_attention"
|
| 74 |
+
],
|
| 75 |
+
"max_position_embeddings": 40960,
|
| 76 |
+
"max_window_layers": 28,
|
| 77 |
+
"model_type": "qwen3",
|
| 78 |
+
"num_attention_heads": 16,
|
| 79 |
+
"num_hidden_layers": 28,
|
| 80 |
+
"num_key_value_heads": 8,
|
| 81 |
+
"output_attentions": false,
|
| 82 |
+
"output_hidden_states": false,
|
| 83 |
+
"pad_token_id": null,
|
| 84 |
+
"problem_type": null,
|
| 85 |
+
"return_dict": true,
|
| 86 |
+
"rms_norm_eps": 1e-06,
|
| 87 |
+
"rope_parameters": {
|
| 88 |
+
"rope_theta": 1000000,
|
| 89 |
+
"rope_type": "default"
|
| 90 |
+
},
|
| 91 |
+
"sliding_window": null,
|
| 92 |
+
"tie_word_embeddings": true,
|
| 93 |
+
"use_cache": true,
|
| 94 |
+
"use_sliding_window": false,
|
| 95 |
+
"vocab_size": 151676
|
| 96 |
+
},
|
| 97 |
+
"model_type": "omnivoice",
|
| 98 |
+
"num_audio_codebook": 8,
|
| 99 |
+
"pad_token_id": 151643,
|
| 100 |
+
"transformers_version": "5.3.0"
|
| 101 |
+
}
|
checkpoint-20000/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:cec5bb9e0e8f3b12449d6ea3f172ffdaf74a3c512dbe2df573f7913741d25eb5
|
| 3 |
+
size 2450344144
|
checkpoint-20000/tokenizer.json
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:408f669b7e2b045fdf54201d815bd364e6667dbd845115da81239c40bc6dcfd1
|
| 3 |
+
size 11423986
|
checkpoint-20000/tokenizer_config.json
ADDED
|
@@ -0,0 +1,23 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": null,
|
| 5 |
+
"clean_up_tokenization_spaces": false,
|
| 6 |
+
"eos_token": "<|im_end|>",
|
| 7 |
+
"errors": "replace",
|
| 8 |
+
"extra_special_tokens": [
|
| 9 |
+
"<|denoise|>",
|
| 10 |
+
"<|lang_start|>",
|
| 11 |
+
"<|lang_end|>",
|
| 12 |
+
"<|instruct_start|>",
|
| 13 |
+
"<|instruct_end|>",
|
| 14 |
+
"<|text_start|>",
|
| 15 |
+
"<|text_end|>"
|
| 16 |
+
],
|
| 17 |
+
"is_local": true,
|
| 18 |
+
"model_max_length": 131072,
|
| 19 |
+
"pad_token": "<|endoftext|>",
|
| 20 |
+
"split_special_tokens": false,
|
| 21 |
+
"tokenizer_class": "Qwen2Tokenizer",
|
| 22 |
+
"unk_token": null
|
| 23 |
+
}
|
checkpoint-20000/train_config.json
ADDED
|
@@ -0,0 +1,58 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"output_dir": "exp/omnivoice_vietnamese_4kh",
|
| 3 |
+
"data_config": "/vast/tts/robert/OmniVoice/data/data_config_vietnamese.json",
|
| 4 |
+
"llm_name_or_path": "Qwen/Qwen3-0.6B",
|
| 5 |
+
"audio_vocab_size": 1025,
|
| 6 |
+
"audio_mask_id": 1024,
|
| 7 |
+
"num_audio_codebook": 8,
|
| 8 |
+
"audio_codebook_weights": [
|
| 9 |
+
8,
|
| 10 |
+
8,
|
| 11 |
+
6,
|
| 12 |
+
6,
|
| 13 |
+
4,
|
| 14 |
+
4,
|
| 15 |
+
2,
|
| 16 |
+
2
|
| 17 |
+
],
|
| 18 |
+
"drop_cond_ratio": 0.1,
|
| 19 |
+
"prompt_ratio_range": [
|
| 20 |
+
0.0,
|
| 21 |
+
0.3
|
| 22 |
+
],
|
| 23 |
+
"mask_ratio_range": [
|
| 24 |
+
0.0,
|
| 25 |
+
1.0
|
| 26 |
+
],
|
| 27 |
+
"language_ratio": 0.8,
|
| 28 |
+
"use_pinyin_ratio": 0.0,
|
| 29 |
+
"instruct_ratio": 0.0,
|
| 30 |
+
"only_instruct_ratio": 0.0,
|
| 31 |
+
"resume_from_checkpoint": null,
|
| 32 |
+
"init_from_checkpoint": "k2-fsa/OmniVoice",
|
| 33 |
+
"learning_rate": 1e-05,
|
| 34 |
+
"weight_decay": 0.01,
|
| 35 |
+
"max_grad_norm": 1.0,
|
| 36 |
+
"steps": 100000,
|
| 37 |
+
"epochs": null,
|
| 38 |
+
"seed": 42,
|
| 39 |
+
"lr_scheduler_type": "cosine",
|
| 40 |
+
"warmup_type": "ratio",
|
| 41 |
+
"warmup_ratio": 0.01,
|
| 42 |
+
"warmup_steps": 0,
|
| 43 |
+
"batch_tokens": 8192,
|
| 44 |
+
"gradient_accumulation_steps": 1,
|
| 45 |
+
"num_workers": 2,
|
| 46 |
+
"mixed_precision": "bf16",
|
| 47 |
+
"allow_tf32": true,
|
| 48 |
+
"use_deepspeed": false,
|
| 49 |
+
"deepspeed_config": null,
|
| 50 |
+
"attn_implementation": "flex_attention",
|
| 51 |
+
"max_sample_tokens": 2000,
|
| 52 |
+
"min_sample_tokens": 50,
|
| 53 |
+
"max_batch_size": 64,
|
| 54 |
+
"logging_steps": 50,
|
| 55 |
+
"eval_steps": 10000,
|
| 56 |
+
"save_steps": 10000,
|
| 57 |
+
"keep_last_n_checkpoints": -1
|
| 58 |
+
}
|
checkpoint-30000/chat_template.jinja
ADDED
|
@@ -0,0 +1,89 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{%- if tools %}
|
| 2 |
+
{{- '<|im_start|>system\n' }}
|
| 3 |
+
{%- if messages[0].role == 'system' %}
|
| 4 |
+
{{- messages[0].content + '\n\n' }}
|
| 5 |
+
{%- endif %}
|
| 6 |
+
{{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
|
| 7 |
+
{%- for tool in tools %}
|
| 8 |
+
{{- "\n" }}
|
| 9 |
+
{{- tool | tojson }}
|
| 10 |
+
{%- endfor %}
|
| 11 |
+
{{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
|
| 12 |
+
{%- else %}
|
| 13 |
+
{%- if messages[0].role == 'system' %}
|
| 14 |
+
{{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
|
| 15 |
+
{%- endif %}
|
| 16 |
+
{%- endif %}
|
| 17 |
+
{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
|
| 18 |
+
{%- for message in messages[::-1] %}
|
| 19 |
+
{%- set index = (messages|length - 1) - loop.index0 %}
|
| 20 |
+
{%- if ns.multi_step_tool and message.role == "user" and message.content is string and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
|
| 21 |
+
{%- set ns.multi_step_tool = false %}
|
| 22 |
+
{%- set ns.last_query_index = index %}
|
| 23 |
+
{%- endif %}
|
| 24 |
+
{%- endfor %}
|
| 25 |
+
{%- for message in messages %}
|
| 26 |
+
{%- if message.content is string %}
|
| 27 |
+
{%- set content = message.content %}
|
| 28 |
+
{%- else %}
|
| 29 |
+
{%- set content = '' %}
|
| 30 |
+
{%- endif %}
|
| 31 |
+
{%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
|
| 32 |
+
{{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
|
| 33 |
+
{%- elif message.role == "assistant" %}
|
| 34 |
+
{%- set reasoning_content = '' %}
|
| 35 |
+
{%- if message.reasoning_content is string %}
|
| 36 |
+
{%- set reasoning_content = message.reasoning_content %}
|
| 37 |
+
{%- else %}
|
| 38 |
+
{%- if '</think>' in content %}
|
| 39 |
+
{%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
|
| 40 |
+
{%- set content = content.split('</think>')[-1].lstrip('\n') %}
|
| 41 |
+
{%- endif %}
|
| 42 |
+
{%- endif %}
|
| 43 |
+
{%- if loop.index0 > ns.last_query_index %}
|
| 44 |
+
{%- if loop.last or (not loop.last and reasoning_content) %}
|
| 45 |
+
{{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
|
| 46 |
+
{%- else %}
|
| 47 |
+
{{- '<|im_start|>' + message.role + '\n' + content }}
|
| 48 |
+
{%- endif %}
|
| 49 |
+
{%- else %}
|
| 50 |
+
{{- '<|im_start|>' + message.role + '\n' + content }}
|
| 51 |
+
{%- endif %}
|
| 52 |
+
{%- if message.tool_calls %}
|
| 53 |
+
{%- for tool_call in message.tool_calls %}
|
| 54 |
+
{%- if (loop.first and content) or (not loop.first) %}
|
| 55 |
+
{{- '\n' }}
|
| 56 |
+
{%- endif %}
|
| 57 |
+
{%- if tool_call.function %}
|
| 58 |
+
{%- set tool_call = tool_call.function %}
|
| 59 |
+
{%- endif %}
|
| 60 |
+
{{- '<tool_call>\n{"name": "' }}
|
| 61 |
+
{{- tool_call.name }}
|
| 62 |
+
{{- '", "arguments": ' }}
|
| 63 |
+
{%- if tool_call.arguments is string %}
|
| 64 |
+
{{- tool_call.arguments }}
|
| 65 |
+
{%- else %}
|
| 66 |
+
{{- tool_call.arguments | tojson }}
|
| 67 |
+
{%- endif %}
|
| 68 |
+
{{- '}\n</tool_call>' }}
|
| 69 |
+
{%- endfor %}
|
| 70 |
+
{%- endif %}
|
| 71 |
+
{{- '<|im_end|>\n' }}
|
| 72 |
+
{%- elif message.role == "tool" %}
|
| 73 |
+
{%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
|
| 74 |
+
{{- '<|im_start|>user' }}
|
| 75 |
+
{%- endif %}
|
| 76 |
+
{{- '\n<tool_response>\n' }}
|
| 77 |
+
{{- content }}
|
| 78 |
+
{{- '\n</tool_response>' }}
|
| 79 |
+
{%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
|
| 80 |
+
{{- '<|im_end|>\n' }}
|
| 81 |
+
{%- endif %}
|
| 82 |
+
{%- endif %}
|
| 83 |
+
{%- endfor %}
|
| 84 |
+
{%- if add_generation_prompt %}
|
| 85 |
+
{{- '<|im_start|>assistant\n' }}
|
| 86 |
+
{%- if enable_thinking is defined and enable_thinking is false %}
|
| 87 |
+
{{- '<think>\n\n</think>\n\n' }}
|
| 88 |
+
{%- endif %}
|
| 89 |
+
{%- endif %}
|
checkpoint-30000/config.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"OmniVoice"
|
| 4 |
+
],
|
| 5 |
+
"audio_codebook_weights": [
|
| 6 |
+
8,
|
| 7 |
+
8,
|
| 8 |
+
6,
|
| 9 |
+
6,
|
| 10 |
+
4,
|
| 11 |
+
4,
|
| 12 |
+
2,
|
| 13 |
+
2
|
| 14 |
+
],
|
| 15 |
+
"audio_mask_id": 1024,
|
| 16 |
+
"audio_vocab_size": 1025,
|
| 17 |
+
"bos_token_id": null,
|
| 18 |
+
"dtype": "float32",
|
| 19 |
+
"eos_token_id": 151645,
|
| 20 |
+
"llm_config": {
|
| 21 |
+
"_name_or_path": "",
|
| 22 |
+
"architectures": [
|
| 23 |
+
"Qwen3ForCausalLM"
|
| 24 |
+
],
|
| 25 |
+
"attention_bias": false,
|
| 26 |
+
"attention_dropout": 0.0,
|
| 27 |
+
"bos_token_id": 151643,
|
| 28 |
+
"chunk_size_feed_forward": 0,
|
| 29 |
+
"dtype": "float32",
|
| 30 |
+
"eos_token_id": 151645,
|
| 31 |
+
"head_dim": 128,
|
| 32 |
+
"hidden_act": "silu",
|
| 33 |
+
"hidden_size": 1024,
|
| 34 |
+
"id2label": {
|
| 35 |
+
"0": "LABEL_0",
|
| 36 |
+
"1": "LABEL_1"
|
| 37 |
+
},
|
| 38 |
+
"initializer_range": 0.02,
|
| 39 |
+
"intermediate_size": 3072,
|
| 40 |
+
"is_encoder_decoder": false,
|
| 41 |
+
"label2id": {
|
| 42 |
+
"LABEL_0": 0,
|
| 43 |
+
"LABEL_1": 1
|
| 44 |
+
},
|
| 45 |
+
"layer_types": [
|
| 46 |
+
"full_attention",
|
| 47 |
+
"full_attention",
|
| 48 |
+
"full_attention",
|
| 49 |
+
"full_attention",
|
| 50 |
+
"full_attention",
|
| 51 |
+
"full_attention",
|
| 52 |
+
"full_attention",
|
| 53 |
+
"full_attention",
|
| 54 |
+
"full_attention",
|
| 55 |
+
"full_attention",
|
| 56 |
+
"full_attention",
|
| 57 |
+
"full_attention",
|
| 58 |
+
"full_attention",
|
| 59 |
+
"full_attention",
|
| 60 |
+
"full_attention",
|
| 61 |
+
"full_attention",
|
| 62 |
+
"full_attention",
|
| 63 |
+
"full_attention",
|
| 64 |
+
"full_attention",
|
| 65 |
+
"full_attention",
|
| 66 |
+
"full_attention",
|
| 67 |
+
"full_attention",
|
| 68 |
+
"full_attention",
|
| 69 |
+
"full_attention",
|
| 70 |
+
"full_attention",
|
| 71 |
+
"full_attention",
|
| 72 |
+
"full_attention",
|
| 73 |
+
"full_attention"
|
| 74 |
+
],
|
| 75 |
+
"max_position_embeddings": 40960,
|
| 76 |
+
"max_window_layers": 28,
|
| 77 |
+
"model_type": "qwen3",
|
| 78 |
+
"num_attention_heads": 16,
|
| 79 |
+
"num_hidden_layers": 28,
|
| 80 |
+
"num_key_value_heads": 8,
|
| 81 |
+
"output_attentions": false,
|
| 82 |
+
"output_hidden_states": false,
|
| 83 |
+
"pad_token_id": null,
|
| 84 |
+
"problem_type": null,
|
| 85 |
+
"return_dict": true,
|
| 86 |
+
"rms_norm_eps": 1e-06,
|
| 87 |
+
"rope_parameters": {
|
| 88 |
+
"rope_theta": 1000000,
|
| 89 |
+
"rope_type": "default"
|
| 90 |
+
},
|
| 91 |
+
"sliding_window": null,
|
| 92 |
+
"tie_word_embeddings": true,
|
| 93 |
+
"use_cache": true,
|
| 94 |
+
"use_sliding_window": false,
|
| 95 |
+
"vocab_size": 151676
|
| 96 |
+
},
|
| 97 |
+
"model_type": "omnivoice",
|
| 98 |
+
"num_audio_codebook": 8,
|
| 99 |
+
"pad_token_id": 151643,
|
| 100 |
+
"transformers_version": "5.3.0"
|
| 101 |
+
}
|
checkpoint-30000/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2ddf1517d7552b03ebe60294daca2077eb71b64e5925e234cc5e453f79b86bea
|
| 3 |
+
size 2450344144
|
checkpoint-30000/tokenizer.json
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:408f669b7e2b045fdf54201d815bd364e6667dbd845115da81239c40bc6dcfd1
|
| 3 |
+
size 11423986
|
checkpoint-30000/tokenizer_config.json
ADDED
|
@@ -0,0 +1,23 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": null,
|
| 5 |
+
"clean_up_tokenization_spaces": false,
|
| 6 |
+
"eos_token": "<|im_end|>",
|
| 7 |
+
"errors": "replace",
|
| 8 |
+
"extra_special_tokens": [
|
| 9 |
+
"<|denoise|>",
|
| 10 |
+
"<|lang_start|>",
|
| 11 |
+
"<|lang_end|>",
|
| 12 |
+
"<|instruct_start|>",
|
| 13 |
+
"<|instruct_end|>",
|
| 14 |
+
"<|text_start|>",
|
| 15 |
+
"<|text_end|>"
|
| 16 |
+
],
|
| 17 |
+
"is_local": true,
|
| 18 |
+
"model_max_length": 131072,
|
| 19 |
+
"pad_token": "<|endoftext|>",
|
| 20 |
+
"split_special_tokens": false,
|
| 21 |
+
"tokenizer_class": "Qwen2Tokenizer",
|
| 22 |
+
"unk_token": null
|
| 23 |
+
}
|
checkpoint-30000/train_config.json
ADDED
|
@@ -0,0 +1,58 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"output_dir": "exp/omnivoice_vietnamese_4kh",
|
| 3 |
+
"data_config": "/vast/tts/robert/OmniVoice/data/data_config_vietnamese.json",
|
| 4 |
+
"llm_name_or_path": "Qwen/Qwen3-0.6B",
|
| 5 |
+
"audio_vocab_size": 1025,
|
| 6 |
+
"audio_mask_id": 1024,
|
| 7 |
+
"num_audio_codebook": 8,
|
| 8 |
+
"audio_codebook_weights": [
|
| 9 |
+
8,
|
| 10 |
+
8,
|
| 11 |
+
6,
|
| 12 |
+
6,
|
| 13 |
+
4,
|
| 14 |
+
4,
|
| 15 |
+
2,
|
| 16 |
+
2
|
| 17 |
+
],
|
| 18 |
+
"drop_cond_ratio": 0.1,
|
| 19 |
+
"prompt_ratio_range": [
|
| 20 |
+
0.0,
|
| 21 |
+
0.3
|
| 22 |
+
],
|
| 23 |
+
"mask_ratio_range": [
|
| 24 |
+
0.0,
|
| 25 |
+
1.0
|
| 26 |
+
],
|
| 27 |
+
"language_ratio": 0.8,
|
| 28 |
+
"use_pinyin_ratio": 0.0,
|
| 29 |
+
"instruct_ratio": 0.0,
|
| 30 |
+
"only_instruct_ratio": 0.0,
|
| 31 |
+
"resume_from_checkpoint": null,
|
| 32 |
+
"init_from_checkpoint": "k2-fsa/OmniVoice",
|
| 33 |
+
"learning_rate": 1e-05,
|
| 34 |
+
"weight_decay": 0.01,
|
| 35 |
+
"max_grad_norm": 1.0,
|
| 36 |
+
"steps": 100000,
|
| 37 |
+
"epochs": null,
|
| 38 |
+
"seed": 42,
|
| 39 |
+
"lr_scheduler_type": "cosine",
|
| 40 |
+
"warmup_type": "ratio",
|
| 41 |
+
"warmup_ratio": 0.01,
|
| 42 |
+
"warmup_steps": 0,
|
| 43 |
+
"batch_tokens": 8192,
|
| 44 |
+
"gradient_accumulation_steps": 1,
|
| 45 |
+
"num_workers": 2,
|
| 46 |
+
"mixed_precision": "bf16",
|
| 47 |
+
"allow_tf32": true,
|
| 48 |
+
"use_deepspeed": false,
|
| 49 |
+
"deepspeed_config": null,
|
| 50 |
+
"attn_implementation": "flex_attention",
|
| 51 |
+
"max_sample_tokens": 2000,
|
| 52 |
+
"min_sample_tokens": 50,
|
| 53 |
+
"max_batch_size": 64,
|
| 54 |
+
"logging_steps": 50,
|
| 55 |
+
"eval_steps": 10000,
|
| 56 |
+
"save_steps": 10000,
|
| 57 |
+
"keep_last_n_checkpoints": -1
|
| 58 |
+
}
|
checkpoint-40000/chat_template.jinja
ADDED
|
@@ -0,0 +1,89 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{%- if tools %}
|
| 2 |
+
{{- '<|im_start|>system\n' }}
|
| 3 |
+
{%- if messages[0].role == 'system' %}
|
| 4 |
+
{{- messages[0].content + '\n\n' }}
|
| 5 |
+
{%- endif %}
|
| 6 |
+
{{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
|
| 7 |
+
{%- for tool in tools %}
|
| 8 |
+
{{- "\n" }}
|
| 9 |
+
{{- tool | tojson }}
|
| 10 |
+
{%- endfor %}
|
| 11 |
+
{{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
|
| 12 |
+
{%- else %}
|
| 13 |
+
{%- if messages[0].role == 'system' %}
|
| 14 |
+
{{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
|
| 15 |
+
{%- endif %}
|
| 16 |
+
{%- endif %}
|
| 17 |
+
{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
|
| 18 |
+
{%- for message in messages[::-1] %}
|
| 19 |
+
{%- set index = (messages|length - 1) - loop.index0 %}
|
| 20 |
+
{%- if ns.multi_step_tool and message.role == "user" and message.content is string and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
|
| 21 |
+
{%- set ns.multi_step_tool = false %}
|
| 22 |
+
{%- set ns.last_query_index = index %}
|
| 23 |
+
{%- endif %}
|
| 24 |
+
{%- endfor %}
|
| 25 |
+
{%- for message in messages %}
|
| 26 |
+
{%- if message.content is string %}
|
| 27 |
+
{%- set content = message.content %}
|
| 28 |
+
{%- else %}
|
| 29 |
+
{%- set content = '' %}
|
| 30 |
+
{%- endif %}
|
| 31 |
+
{%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
|
| 32 |
+
{{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
|
| 33 |
+
{%- elif message.role == "assistant" %}
|
| 34 |
+
{%- set reasoning_content = '' %}
|
| 35 |
+
{%- if message.reasoning_content is string %}
|
| 36 |
+
{%- set reasoning_content = message.reasoning_content %}
|
| 37 |
+
{%- else %}
|
| 38 |
+
{%- if '</think>' in content %}
|
| 39 |
+
{%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
|
| 40 |
+
{%- set content = content.split('</think>')[-1].lstrip('\n') %}
|
| 41 |
+
{%- endif %}
|
| 42 |
+
{%- endif %}
|
| 43 |
+
{%- if loop.index0 > ns.last_query_index %}
|
| 44 |
+
{%- if loop.last or (not loop.last and reasoning_content) %}
|
| 45 |
+
{{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
|
| 46 |
+
{%- else %}
|
| 47 |
+
{{- '<|im_start|>' + message.role + '\n' + content }}
|
| 48 |
+
{%- endif %}
|
| 49 |
+
{%- else %}
|
| 50 |
+
{{- '<|im_start|>' + message.role + '\n' + content }}
|
| 51 |
+
{%- endif %}
|
| 52 |
+
{%- if message.tool_calls %}
|
| 53 |
+
{%- for tool_call in message.tool_calls %}
|
| 54 |
+
{%- if (loop.first and content) or (not loop.first) %}
|
| 55 |
+
{{- '\n' }}
|
| 56 |
+
{%- endif %}
|
| 57 |
+
{%- if tool_call.function %}
|
| 58 |
+
{%- set tool_call = tool_call.function %}
|
| 59 |
+
{%- endif %}
|
| 60 |
+
{{- '<tool_call>\n{"name": "' }}
|
| 61 |
+
{{- tool_call.name }}
|
| 62 |
+
{{- '", "arguments": ' }}
|
| 63 |
+
{%- if tool_call.arguments is string %}
|
| 64 |
+
{{- tool_call.arguments }}
|
| 65 |
+
{%- else %}
|
| 66 |
+
{{- tool_call.arguments | tojson }}
|
| 67 |
+
{%- endif %}
|
| 68 |
+
{{- '}\n</tool_call>' }}
|
| 69 |
+
{%- endfor %}
|
| 70 |
+
{%- endif %}
|
| 71 |
+
{{- '<|im_end|>\n' }}
|
| 72 |
+
{%- elif message.role == "tool" %}
|
| 73 |
+
{%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
|
| 74 |
+
{{- '<|im_start|>user' }}
|
| 75 |
+
{%- endif %}
|
| 76 |
+
{{- '\n<tool_response>\n' }}
|
| 77 |
+
{{- content }}
|
| 78 |
+
{{- '\n</tool_response>' }}
|
| 79 |
+
{%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
|
| 80 |
+
{{- '<|im_end|>\n' }}
|
| 81 |
+
{%- endif %}
|
| 82 |
+
{%- endif %}
|
| 83 |
+
{%- endfor %}
|
| 84 |
+
{%- if add_generation_prompt %}
|
| 85 |
+
{{- '<|im_start|>assistant\n' }}
|
| 86 |
+
{%- if enable_thinking is defined and enable_thinking is false %}
|
| 87 |
+
{{- '<think>\n\n</think>\n\n' }}
|
| 88 |
+
{%- endif %}
|
| 89 |
+
{%- endif %}
|
checkpoint-40000/config.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"OmniVoice"
|
| 4 |
+
],
|
| 5 |
+
"audio_codebook_weights": [
|
| 6 |
+
8,
|
| 7 |
+
8,
|
| 8 |
+
6,
|
| 9 |
+
6,
|
| 10 |
+
4,
|
| 11 |
+
4,
|
| 12 |
+
2,
|
| 13 |
+
2
|
| 14 |
+
],
|
| 15 |
+
"audio_mask_id": 1024,
|
| 16 |
+
"audio_vocab_size": 1025,
|
| 17 |
+
"bos_token_id": null,
|
| 18 |
+
"dtype": "float32",
|
| 19 |
+
"eos_token_id": 151645,
|
| 20 |
+
"llm_config": {
|
| 21 |
+
"_name_or_path": "",
|
| 22 |
+
"architectures": [
|
| 23 |
+
"Qwen3ForCausalLM"
|
| 24 |
+
],
|
| 25 |
+
"attention_bias": false,
|
| 26 |
+
"attention_dropout": 0.0,
|
| 27 |
+
"bos_token_id": 151643,
|
| 28 |
+
"chunk_size_feed_forward": 0,
|
| 29 |
+
"dtype": "float32",
|
| 30 |
+
"eos_token_id": 151645,
|
| 31 |
+
"head_dim": 128,
|
| 32 |
+
"hidden_act": "silu",
|
| 33 |
+
"hidden_size": 1024,
|
| 34 |
+
"id2label": {
|
| 35 |
+
"0": "LABEL_0",
|
| 36 |
+
"1": "LABEL_1"
|
| 37 |
+
},
|
| 38 |
+
"initializer_range": 0.02,
|
| 39 |
+
"intermediate_size": 3072,
|
| 40 |
+
"is_encoder_decoder": false,
|
| 41 |
+
"label2id": {
|
| 42 |
+
"LABEL_0": 0,
|
| 43 |
+
"LABEL_1": 1
|
| 44 |
+
},
|
| 45 |
+
"layer_types": [
|
| 46 |
+
"full_attention",
|
| 47 |
+
"full_attention",
|
| 48 |
+
"full_attention",
|
| 49 |
+
"full_attention",
|
| 50 |
+
"full_attention",
|
| 51 |
+
"full_attention",
|
| 52 |
+
"full_attention",
|
| 53 |
+
"full_attention",
|
| 54 |
+
"full_attention",
|
| 55 |
+
"full_attention",
|
| 56 |
+
"full_attention",
|
| 57 |
+
"full_attention",
|
| 58 |
+
"full_attention",
|
| 59 |
+
"full_attention",
|
| 60 |
+
"full_attention",
|
| 61 |
+
"full_attention",
|
| 62 |
+
"full_attention",
|
| 63 |
+
"full_attention",
|
| 64 |
+
"full_attention",
|
| 65 |
+
"full_attention",
|
| 66 |
+
"full_attention",
|
| 67 |
+
"full_attention",
|
| 68 |
+
"full_attention",
|
| 69 |
+
"full_attention",
|
| 70 |
+
"full_attention",
|
| 71 |
+
"full_attention",
|
| 72 |
+
"full_attention",
|
| 73 |
+
"full_attention"
|
| 74 |
+
],
|
| 75 |
+
"max_position_embeddings": 40960,
|
| 76 |
+
"max_window_layers": 28,
|
| 77 |
+
"model_type": "qwen3",
|
| 78 |
+
"num_attention_heads": 16,
|
| 79 |
+
"num_hidden_layers": 28,
|
| 80 |
+
"num_key_value_heads": 8,
|
| 81 |
+
"output_attentions": false,
|
| 82 |
+
"output_hidden_states": false,
|
| 83 |
+
"pad_token_id": null,
|
| 84 |
+
"problem_type": null,
|
| 85 |
+
"return_dict": true,
|
| 86 |
+
"rms_norm_eps": 1e-06,
|
| 87 |
+
"rope_parameters": {
|
| 88 |
+
"rope_theta": 1000000,
|
| 89 |
+
"rope_type": "default"
|
| 90 |
+
},
|
| 91 |
+
"sliding_window": null,
|
| 92 |
+
"tie_word_embeddings": true,
|
| 93 |
+
"use_cache": true,
|
| 94 |
+
"use_sliding_window": false,
|
| 95 |
+
"vocab_size": 151676
|
| 96 |
+
},
|
| 97 |
+
"model_type": "omnivoice",
|
| 98 |
+
"num_audio_codebook": 8,
|
| 99 |
+
"pad_token_id": 151643,
|
| 100 |
+
"transformers_version": "5.3.0"
|
| 101 |
+
}
|
checkpoint-40000/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:6942079aed66b18552dc1ee3fd511f7156f25163281056a3885e1aa41a44a45c
|
| 3 |
+
size 2450344144
|
checkpoint-40000/tokenizer.json
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:408f669b7e2b045fdf54201d815bd364e6667dbd845115da81239c40bc6dcfd1
|
| 3 |
+
size 11423986
|
checkpoint-40000/tokenizer_config.json
ADDED
|
@@ -0,0 +1,23 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": null,
|
| 5 |
+
"clean_up_tokenization_spaces": false,
|
| 6 |
+
"eos_token": "<|im_end|>",
|
| 7 |
+
"errors": "replace",
|
| 8 |
+
"extra_special_tokens": [
|
| 9 |
+
"<|denoise|>",
|
| 10 |
+
"<|lang_start|>",
|
| 11 |
+
"<|lang_end|>",
|
| 12 |
+
"<|instruct_start|>",
|
| 13 |
+
"<|instruct_end|>",
|
| 14 |
+
"<|text_start|>",
|
| 15 |
+
"<|text_end|>"
|
| 16 |
+
],
|
| 17 |
+
"is_local": true,
|
| 18 |
+
"model_max_length": 131072,
|
| 19 |
+
"pad_token": "<|endoftext|>",
|
| 20 |
+
"split_special_tokens": false,
|
| 21 |
+
"tokenizer_class": "Qwen2Tokenizer",
|
| 22 |
+
"unk_token": null
|
| 23 |
+
}
|
checkpoint-40000/train_config.json
ADDED
|
@@ -0,0 +1,58 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"output_dir": "exp/omnivoice_vietnamese_4kh",
|
| 3 |
+
"data_config": "/vast/tts/robert/OmniVoice/data/data_config_vietnamese.json",
|
| 4 |
+
"llm_name_or_path": "Qwen/Qwen3-0.6B",
|
| 5 |
+
"audio_vocab_size": 1025,
|
| 6 |
+
"audio_mask_id": 1024,
|
| 7 |
+
"num_audio_codebook": 8,
|
| 8 |
+
"audio_codebook_weights": [
|
| 9 |
+
8,
|
| 10 |
+
8,
|
| 11 |
+
6,
|
| 12 |
+
6,
|
| 13 |
+
4,
|
| 14 |
+
4,
|
| 15 |
+
2,
|
| 16 |
+
2
|
| 17 |
+
],
|
| 18 |
+
"drop_cond_ratio": 0.1,
|
| 19 |
+
"prompt_ratio_range": [
|
| 20 |
+
0.0,
|
| 21 |
+
0.3
|
| 22 |
+
],
|
| 23 |
+
"mask_ratio_range": [
|
| 24 |
+
0.0,
|
| 25 |
+
1.0
|
| 26 |
+
],
|
| 27 |
+
"language_ratio": 0.8,
|
| 28 |
+
"use_pinyin_ratio": 0.0,
|
| 29 |
+
"instruct_ratio": 0.0,
|
| 30 |
+
"only_instruct_ratio": 0.0,
|
| 31 |
+
"resume_from_checkpoint": null,
|
| 32 |
+
"init_from_checkpoint": "k2-fsa/OmniVoice",
|
| 33 |
+
"learning_rate": 1e-05,
|
| 34 |
+
"weight_decay": 0.01,
|
| 35 |
+
"max_grad_norm": 1.0,
|
| 36 |
+
"steps": 100000,
|
| 37 |
+
"epochs": null,
|
| 38 |
+
"seed": 42,
|
| 39 |
+
"lr_scheduler_type": "cosine",
|
| 40 |
+
"warmup_type": "ratio",
|
| 41 |
+
"warmup_ratio": 0.01,
|
| 42 |
+
"warmup_steps": 0,
|
| 43 |
+
"batch_tokens": 8192,
|
| 44 |
+
"gradient_accumulation_steps": 1,
|
| 45 |
+
"num_workers": 2,
|
| 46 |
+
"mixed_precision": "bf16",
|
| 47 |
+
"allow_tf32": true,
|
| 48 |
+
"use_deepspeed": false,
|
| 49 |
+
"deepspeed_config": null,
|
| 50 |
+
"attn_implementation": "flex_attention",
|
| 51 |
+
"max_sample_tokens": 2000,
|
| 52 |
+
"min_sample_tokens": 50,
|
| 53 |
+
"max_batch_size": 64,
|
| 54 |
+
"logging_steps": 50,
|
| 55 |
+
"eval_steps": 10000,
|
| 56 |
+
"save_steps": 10000,
|
| 57 |
+
"keep_last_n_checkpoints": -1
|
| 58 |
+
}
|
checkpoint-50000/chat_template.jinja
ADDED
|
@@ -0,0 +1,89 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{%- if tools %}
|
| 2 |
+
{{- '<|im_start|>system\n' }}
|
| 3 |
+
{%- if messages[0].role == 'system' %}
|
| 4 |
+
{{- messages[0].content + '\n\n' }}
|
| 5 |
+
{%- endif %}
|
| 6 |
+
{{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
|
| 7 |
+
{%- for tool in tools %}
|
| 8 |
+
{{- "\n" }}
|
| 9 |
+
{{- tool | tojson }}
|
| 10 |
+
{%- endfor %}
|
| 11 |
+
{{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
|
| 12 |
+
{%- else %}
|
| 13 |
+
{%- if messages[0].role == 'system' %}
|
| 14 |
+
{{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
|
| 15 |
+
{%- endif %}
|
| 16 |
+
{%- endif %}
|
| 17 |
+
{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
|
| 18 |
+
{%- for message in messages[::-1] %}
|
| 19 |
+
{%- set index = (messages|length - 1) - loop.index0 %}
|
| 20 |
+
{%- if ns.multi_step_tool and message.role == "user" and message.content is string and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
|
| 21 |
+
{%- set ns.multi_step_tool = false %}
|
| 22 |
+
{%- set ns.last_query_index = index %}
|
| 23 |
+
{%- endif %}
|
| 24 |
+
{%- endfor %}
|
| 25 |
+
{%- for message in messages %}
|
| 26 |
+
{%- if message.content is string %}
|
| 27 |
+
{%- set content = message.content %}
|
| 28 |
+
{%- else %}
|
| 29 |
+
{%- set content = '' %}
|
| 30 |
+
{%- endif %}
|
| 31 |
+
{%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
|
| 32 |
+
{{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
|
| 33 |
+
{%- elif message.role == "assistant" %}
|
| 34 |
+
{%- set reasoning_content = '' %}
|
| 35 |
+
{%- if message.reasoning_content is string %}
|
| 36 |
+
{%- set reasoning_content = message.reasoning_content %}
|
| 37 |
+
{%- else %}
|
| 38 |
+
{%- if '</think>' in content %}
|
| 39 |
+
{%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
|
| 40 |
+
{%- set content = content.split('</think>')[-1].lstrip('\n') %}
|
| 41 |
+
{%- endif %}
|
| 42 |
+
{%- endif %}
|
| 43 |
+
{%- if loop.index0 > ns.last_query_index %}
|
| 44 |
+
{%- if loop.last or (not loop.last and reasoning_content) %}
|
| 45 |
+
{{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
|
| 46 |
+
{%- else %}
|
| 47 |
+
{{- '<|im_start|>' + message.role + '\n' + content }}
|
| 48 |
+
{%- endif %}
|
| 49 |
+
{%- else %}
|
| 50 |
+
{{- '<|im_start|>' + message.role + '\n' + content }}
|
| 51 |
+
{%- endif %}
|
| 52 |
+
{%- if message.tool_calls %}
|
| 53 |
+
{%- for tool_call in message.tool_calls %}
|
| 54 |
+
{%- if (loop.first and content) or (not loop.first) %}
|
| 55 |
+
{{- '\n' }}
|
| 56 |
+
{%- endif %}
|
| 57 |
+
{%- if tool_call.function %}
|
| 58 |
+
{%- set tool_call = tool_call.function %}
|
| 59 |
+
{%- endif %}
|
| 60 |
+
{{- '<tool_call>\n{"name": "' }}
|
| 61 |
+
{{- tool_call.name }}
|
| 62 |
+
{{- '", "arguments": ' }}
|
| 63 |
+
{%- if tool_call.arguments is string %}
|
| 64 |
+
{{- tool_call.arguments }}
|
| 65 |
+
{%- else %}
|
| 66 |
+
{{- tool_call.arguments | tojson }}
|
| 67 |
+
{%- endif %}
|
| 68 |
+
{{- '}\n</tool_call>' }}
|
| 69 |
+
{%- endfor %}
|
| 70 |
+
{%- endif %}
|
| 71 |
+
{{- '<|im_end|>\n' }}
|
| 72 |
+
{%- elif message.role == "tool" %}
|
| 73 |
+
{%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
|
| 74 |
+
{{- '<|im_start|>user' }}
|
| 75 |
+
{%- endif %}
|
| 76 |
+
{{- '\n<tool_response>\n' }}
|
| 77 |
+
{{- content }}
|
| 78 |
+
{{- '\n</tool_response>' }}
|
| 79 |
+
{%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
|
| 80 |
+
{{- '<|im_end|>\n' }}
|
| 81 |
+
{%- endif %}
|
| 82 |
+
{%- endif %}
|
| 83 |
+
{%- endfor %}
|
| 84 |
+
{%- if add_generation_prompt %}
|
| 85 |
+
{{- '<|im_start|>assistant\n' }}
|
| 86 |
+
{%- if enable_thinking is defined and enable_thinking is false %}
|
| 87 |
+
{{- '<think>\n\n</think>\n\n' }}
|
| 88 |
+
{%- endif %}
|
| 89 |
+
{%- endif %}
|
checkpoint-50000/config.json
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"OmniVoice"
|
| 4 |
+
],
|
| 5 |
+
"audio_codebook_weights": [
|
| 6 |
+
8,
|
| 7 |
+
8,
|
| 8 |
+
6,
|
| 9 |
+
6,
|
| 10 |
+
4,
|
| 11 |
+
4,
|
| 12 |
+
2,
|
| 13 |
+
2
|
| 14 |
+
],
|
| 15 |
+
"audio_mask_id": 1024,
|
| 16 |
+
"audio_vocab_size": 1025,
|
| 17 |
+
"bos_token_id": null,
|
| 18 |
+
"dtype": "float32",
|
| 19 |
+
"eos_token_id": 151645,
|
| 20 |
+
"llm_config": {
|
| 21 |
+
"_name_or_path": "",
|
| 22 |
+
"architectures": [
|
| 23 |
+
"Qwen3ForCausalLM"
|
| 24 |
+
],
|
| 25 |
+
"attention_bias": false,
|
| 26 |
+
"attention_dropout": 0.0,
|
| 27 |
+
"bos_token_id": 151643,
|
| 28 |
+
"chunk_size_feed_forward": 0,
|
| 29 |
+
"dtype": "float32",
|
| 30 |
+
"eos_token_id": 151645,
|
| 31 |
+
"head_dim": 128,
|
| 32 |
+
"hidden_act": "silu",
|
| 33 |
+
"hidden_size": 1024,
|
| 34 |
+
"id2label": {
|
| 35 |
+
"0": "LABEL_0",
|
| 36 |
+
"1": "LABEL_1"
|
| 37 |
+
},
|
| 38 |
+
"initializer_range": 0.02,
|
| 39 |
+
"intermediate_size": 3072,
|
| 40 |
+
"is_encoder_decoder": false,
|
| 41 |
+
"label2id": {
|
| 42 |
+
"LABEL_0": 0,
|
| 43 |
+
"LABEL_1": 1
|
| 44 |
+
},
|
| 45 |
+
"layer_types": [
|
| 46 |
+
"full_attention",
|
| 47 |
+
"full_attention",
|
| 48 |
+
"full_attention",
|
| 49 |
+
"full_attention",
|
| 50 |
+
"full_attention",
|
| 51 |
+
"full_attention",
|
| 52 |
+
"full_attention",
|
| 53 |
+
"full_attention",
|
| 54 |
+
"full_attention",
|
| 55 |
+
"full_attention",
|
| 56 |
+
"full_attention",
|
| 57 |
+
"full_attention",
|
| 58 |
+
"full_attention",
|
| 59 |
+
"full_attention",
|
| 60 |
+
"full_attention",
|
| 61 |
+
"full_attention",
|
| 62 |
+
"full_attention",
|
| 63 |
+
"full_attention",
|
| 64 |
+
"full_attention",
|
| 65 |
+
"full_attention",
|
| 66 |
+
"full_attention",
|
| 67 |
+
"full_attention",
|
| 68 |
+
"full_attention",
|
| 69 |
+
"full_attention",
|
| 70 |
+
"full_attention",
|
| 71 |
+
"full_attention",
|
| 72 |
+
"full_attention",
|
| 73 |
+
"full_attention"
|
| 74 |
+
],
|
| 75 |
+
"max_position_embeddings": 40960,
|
| 76 |
+
"max_window_layers": 28,
|
| 77 |
+
"model_type": "qwen3",
|
| 78 |
+
"num_attention_heads": 16,
|
| 79 |
+
"num_hidden_layers": 28,
|
| 80 |
+
"num_key_value_heads": 8,
|
| 81 |
+
"output_attentions": false,
|
| 82 |
+
"output_hidden_states": false,
|
| 83 |
+
"pad_token_id": null,
|
| 84 |
+
"problem_type": null,
|
| 85 |
+
"return_dict": true,
|
| 86 |
+
"rms_norm_eps": 1e-06,
|
| 87 |
+
"rope_parameters": {
|
| 88 |
+
"rope_theta": 1000000,
|
| 89 |
+
"rope_type": "default"
|
| 90 |
+
},
|
| 91 |
+
"sliding_window": null,
|
| 92 |
+
"tie_word_embeddings": true,
|
| 93 |
+
"use_cache": true,
|
| 94 |
+
"use_sliding_window": false,
|
| 95 |
+
"vocab_size": 151676
|
| 96 |
+
},
|
| 97 |
+
"model_type": "omnivoice",
|
| 98 |
+
"num_audio_codebook": 8,
|
| 99 |
+
"pad_token_id": 151643,
|
| 100 |
+
"transformers_version": "5.3.0"
|
| 101 |
+
}
|
checkpoint-50000/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:4b2e1563aac400e06ad86fb62c4f9a759d3f81ae0b662c6afbc9aeb19607ce20
|
| 3 |
+
size 2450344144
|
checkpoint-50000/tokenizer.json
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:408f669b7e2b045fdf54201d815bd364e6667dbd845115da81239c40bc6dcfd1
|
| 3 |
+
size 11423986
|
checkpoint-50000/tokenizer_config.json
ADDED
|
@@ -0,0 +1,23 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": null,
|
| 5 |
+
"clean_up_tokenization_spaces": false,
|
| 6 |
+
"eos_token": "<|im_end|>",
|
| 7 |
+
"errors": "replace",
|
| 8 |
+
"extra_special_tokens": [
|
| 9 |
+
"<|denoise|>",
|
| 10 |
+
"<|lang_start|>",
|
| 11 |
+
"<|lang_end|>",
|
| 12 |
+
"<|instruct_start|>",
|
| 13 |
+
"<|instruct_end|>",
|
| 14 |
+
"<|text_start|>",
|
| 15 |
+
"<|text_end|>"
|
| 16 |
+
],
|
| 17 |
+
"is_local": true,
|
| 18 |
+
"model_max_length": 131072,
|
| 19 |
+
"pad_token": "<|endoftext|>",
|
| 20 |
+
"split_special_tokens": false,
|
| 21 |
+
"tokenizer_class": "Qwen2Tokenizer",
|
| 22 |
+
"unk_token": null
|
| 23 |
+
}
|
checkpoint-50000/train_config.json
ADDED
|
@@ -0,0 +1,58 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"output_dir": "exp/omnivoice_vietnamese_4kh",
|
| 3 |
+
"data_config": "/vast/tts/robert/OmniVoice/data/data_config_vietnamese.json",
|
| 4 |
+
"llm_name_or_path": "Qwen/Qwen3-0.6B",
|
| 5 |
+
"audio_vocab_size": 1025,
|
| 6 |
+
"audio_mask_id": 1024,
|
| 7 |
+
"num_audio_codebook": 8,
|
| 8 |
+
"audio_codebook_weights": [
|
| 9 |
+
8,
|
| 10 |
+
8,
|
| 11 |
+
6,
|
| 12 |
+
6,
|
| 13 |
+
4,
|
| 14 |
+
4,
|
| 15 |
+
2,
|
| 16 |
+
2
|
| 17 |
+
],
|
| 18 |
+
"drop_cond_ratio": 0.1,
|
| 19 |
+
"prompt_ratio_range": [
|
| 20 |
+
0.0,
|
| 21 |
+
0.3
|
| 22 |
+
],
|
| 23 |
+
"mask_ratio_range": [
|
| 24 |
+
0.0,
|
| 25 |
+
1.0
|
| 26 |
+
],
|
| 27 |
+
"language_ratio": 0.8,
|
| 28 |
+
"use_pinyin_ratio": 0.0,
|
| 29 |
+
"instruct_ratio": 0.0,
|
| 30 |
+
"only_instruct_ratio": 0.0,
|
| 31 |
+
"resume_from_checkpoint": null,
|
| 32 |
+
"init_from_checkpoint": "k2-fsa/OmniVoice",
|
| 33 |
+
"learning_rate": 1e-05,
|
| 34 |
+
"weight_decay": 0.01,
|
| 35 |
+
"max_grad_norm": 1.0,
|
| 36 |
+
"steps": 100000,
|
| 37 |
+
"epochs": null,
|
| 38 |
+
"seed": 42,
|
| 39 |
+
"lr_scheduler_type": "cosine",
|
| 40 |
+
"warmup_type": "ratio",
|
| 41 |
+
"warmup_ratio": 0.01,
|
| 42 |
+
"warmup_steps": 0,
|
| 43 |
+
"batch_tokens": 8192,
|
| 44 |
+
"gradient_accumulation_steps": 1,
|
| 45 |
+
"num_workers": 2,
|
| 46 |
+
"mixed_precision": "bf16",
|
| 47 |
+
"allow_tf32": true,
|
| 48 |
+
"use_deepspeed": false,
|
| 49 |
+
"deepspeed_config": null,
|
| 50 |
+
"attn_implementation": "flex_attention",
|
| 51 |
+
"max_sample_tokens": 2000,
|
| 52 |
+
"min_sample_tokens": 50,
|
| 53 |
+
"max_batch_size": 64,
|
| 54 |
+
"logging_steps": 50,
|
| 55 |
+
"eval_steps": 10000,
|
| 56 |
+
"save_steps": 10000,
|
| 57 |
+
"keep_last_n_checkpoints": -1
|
| 58 |
+
}
|
initial_config.json
ADDED
|
@@ -0,0 +1,58 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"output_dir": "exp/omnivoice_vietnamese_4kh",
|
| 3 |
+
"data_config": "/vast/tts/robert/OmniVoice/data/data_config_vietnamese.json",
|
| 4 |
+
"llm_name_or_path": "Qwen/Qwen3-0.6B",
|
| 5 |
+
"audio_vocab_size": 1025,
|
| 6 |
+
"audio_mask_id": 1024,
|
| 7 |
+
"num_audio_codebook": 8,
|
| 8 |
+
"audio_codebook_weights": [
|
| 9 |
+
8,
|
| 10 |
+
8,
|
| 11 |
+
6,
|
| 12 |
+
6,
|
| 13 |
+
4,
|
| 14 |
+
4,
|
| 15 |
+
2,
|
| 16 |
+
2
|
| 17 |
+
],
|
| 18 |
+
"drop_cond_ratio": 0.1,
|
| 19 |
+
"prompt_ratio_range": [
|
| 20 |
+
0.0,
|
| 21 |
+
0.3
|
| 22 |
+
],
|
| 23 |
+
"mask_ratio_range": [
|
| 24 |
+
0.0,
|
| 25 |
+
1.0
|
| 26 |
+
],
|
| 27 |
+
"language_ratio": 0.8,
|
| 28 |
+
"use_pinyin_ratio": 0.0,
|
| 29 |
+
"instruct_ratio": 0.0,
|
| 30 |
+
"only_instruct_ratio": 0.0,
|
| 31 |
+
"resume_from_checkpoint": null,
|
| 32 |
+
"init_from_checkpoint": "k2-fsa/OmniVoice",
|
| 33 |
+
"learning_rate": 1e-05,
|
| 34 |
+
"weight_decay": 0.01,
|
| 35 |
+
"max_grad_norm": 1.0,
|
| 36 |
+
"steps": 100000,
|
| 37 |
+
"epochs": null,
|
| 38 |
+
"seed": 42,
|
| 39 |
+
"lr_scheduler_type": "cosine",
|
| 40 |
+
"warmup_type": "ratio",
|
| 41 |
+
"warmup_ratio": 0.01,
|
| 42 |
+
"warmup_steps": 0,
|
| 43 |
+
"batch_tokens": 8192,
|
| 44 |
+
"gradient_accumulation_steps": 1,
|
| 45 |
+
"num_workers": 2,
|
| 46 |
+
"mixed_precision": "bf16",
|
| 47 |
+
"allow_tf32": true,
|
| 48 |
+
"use_deepspeed": false,
|
| 49 |
+
"deepspeed_config": null,
|
| 50 |
+
"attn_implementation": "flex_attention",
|
| 51 |
+
"max_sample_tokens": 2000,
|
| 52 |
+
"min_sample_tokens": 50,
|
| 53 |
+
"max_batch_size": 64,
|
| 54 |
+
"logging_steps": 50,
|
| 55 |
+
"eval_steps": 10000,
|
| 56 |
+
"save_steps": 10000,
|
| 57 |
+
"keep_last_n_checkpoints": -1
|
| 58 |
+
}
|
tensorboard/events.out.tfevents.1777886498.osaka.2227663.0
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:33b52543081639f348aa73710bccf6852e984a8b663c7e8e4f18fdb93b945add
|
| 3 |
+
size 317849
|
train.log
ADDED
|
@@ -0,0 +1,50 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
05/04/2026 09:03:47 - INFO - omnivoice.training.trainer - Loaded Config: TrainingConfig(output_dir='exp/omnivoice_vietnamese_4kh', data_config='/vast/tts/robert/OmniVoice/data/data_config_vietnamese.json', llm_name_or_path='Qwen/Qwen3-0.6B', audio_vocab_size=1025, audio_mask_id=1024, num_audio_codebook=8, audio_codebook_weights=[8, 8, 6, 6, 4, 4, 2, 2], drop_cond_ratio=0.1, prompt_ratio_range=[0.0, 0.3], mask_ratio_range=[0.0, 1.0], language_ratio=0.8, use_pinyin_ratio=0.0, instruct_ratio=0.0, only_instruct_ratio=0.0, resume_from_checkpoint=None, init_from_checkpoint='k2-fsa/OmniVoice', learning_rate=1e-05, weight_decay=0.01, max_grad_norm=1.0, steps=300000, epochs=2, seed=42, lr_scheduler_type='cosine', warmup_type='ratio', warmup_ratio=0.01, warmup_steps=0, batch_tokens=8192, gradient_accumulation_steps=1, num_workers=2, mixed_precision='bf16', allow_tf32=True, use_deepspeed=False, deepspeed_config=None, attn_implementation='flex_attention', max_sample_tokens=2000, min_sample_tokens=50, max_batch_size=64, logging_steps=50, eval_steps=10000, save_steps=10000, keep_last_n_checkpoints=-1)
|
| 2 |
+
05/04/2026 09:21:34 - INFO - omnivoice.training.trainer - Loaded Config: TrainingConfig(output_dir='exp/omnivoice_vietnamese_4kh', data_config='/vast/tts/robert/OmniVoice/data/data_config_vietnamese.json', llm_name_or_path='Qwen/Qwen3-0.6B', audio_vocab_size=1025, audio_mask_id=1024, num_audio_codebook=8, audio_codebook_weights=[8, 8, 6, 6, 4, 4, 2, 2], drop_cond_ratio=0.1, prompt_ratio_range=[0.0, 0.3], mask_ratio_range=[0.0, 1.0], language_ratio=0.8, use_pinyin_ratio=0.0, instruct_ratio=0.0, only_instruct_ratio=0.0, resume_from_checkpoint=None, init_from_checkpoint='k2-fsa/OmniVoice', learning_rate=1e-05, weight_decay=0.01, max_grad_norm=1.0, steps=100000, epochs=None, seed=42, lr_scheduler_type='cosine', warmup_type='ratio', warmup_ratio=0.01, warmup_steps=0, batch_tokens=8192, gradient_accumulation_steps=1, num_workers=2, mixed_precision='bf16', allow_tf32=True, use_deepspeed=False, deepspeed_config=None, attn_implementation='flex_attention', max_sample_tokens=2000, min_sample_tokens=50, max_batch_size=64, logging_steps=50, eval_steps=10000, save_steps=10000, keep_last_n_checkpoints=-1)
|
| 3 |
+
05/04/2026 09:21:40 - INFO - omnivoice.training.trainer - Starting Training Loop...
|
| 4 |
+
05/04/2026 10:07:11 - INFO - omnivoice.training.trainer - Epoch 1 starting. Resetting dataloader...
|
| 5 |
+
05/04/2026 10:18:44 - INFO - omnivoice.training.trainer - Running evaluation at step 10000...
|
| 6 |
+
05/04/2026 10:18:47 - INFO - omnivoice.training.trainer - Eval Loss: 4.1190
|
| 7 |
+
05/04/2026 10:18:47 - INFO - accelerate.accelerator - Saving current state to exp/omnivoice_vietnamese_4kh/checkpoint-10000
|
| 8 |
+
05/04/2026 10:18:57 - INFO - accelerate.checkpointing - Model weights saved in exp/omnivoice_vietnamese_4kh/checkpoint-10000/model.safetensors
|
| 9 |
+
05/04/2026 10:19:02 - INFO - accelerate.checkpointing - Optimizer state saved in exp/omnivoice_vietnamese_4kh/checkpoint-10000/optimizer.bin
|
| 10 |
+
05/04/2026 10:19:02 - INFO - accelerate.checkpointing - Scheduler state saved in exp/omnivoice_vietnamese_4kh/checkpoint-10000/scheduler.bin
|
| 11 |
+
05/04/2026 10:19:02 - INFO - accelerate.checkpointing - Random states saved in exp/omnivoice_vietnamese_4kh/checkpoint-10000/random_states_0.pkl
|
| 12 |
+
05/04/2026 10:19:05 - INFO - omnivoice.training.checkpoint - Saved checkpoint to exp/omnivoice_vietnamese_4kh/checkpoint-10000
|
| 13 |
+
05/04/2026 10:52:54 - INFO - omnivoice.training.trainer - Epoch 2 starting. Resetting dataloader...
|
| 14 |
+
05/04/2026 11:15:51 - INFO - omnivoice.training.trainer - Running evaluation at step 20000...
|
| 15 |
+
05/04/2026 11:15:53 - INFO - omnivoice.training.trainer - Eval Loss: 4.1431
|
| 16 |
+
05/04/2026 11:15:53 - INFO - accelerate.accelerator - Saving current state to exp/omnivoice_vietnamese_4kh/checkpoint-20000
|
| 17 |
+
05/04/2026 11:16:02 - INFO - accelerate.checkpointing - Model weights saved in exp/omnivoice_vietnamese_4kh/checkpoint-20000/model.safetensors
|
| 18 |
+
05/04/2026 11:16:07 - INFO - accelerate.checkpointing - Optimizer state saved in exp/omnivoice_vietnamese_4kh/checkpoint-20000/optimizer.bin
|
| 19 |
+
05/04/2026 11:16:07 - INFO - accelerate.checkpointing - Scheduler state saved in exp/omnivoice_vietnamese_4kh/checkpoint-20000/scheduler.bin
|
| 20 |
+
05/04/2026 11:16:07 - INFO - accelerate.checkpointing - Random states saved in exp/omnivoice_vietnamese_4kh/checkpoint-20000/random_states_0.pkl
|
| 21 |
+
05/04/2026 11:16:09 - INFO - omnivoice.training.checkpoint - Saved checkpoint to exp/omnivoice_vietnamese_4kh/checkpoint-20000
|
| 22 |
+
05/04/2026 11:38:34 - INFO - omnivoice.training.trainer - Epoch 3 starting. Resetting dataloader...
|
| 23 |
+
05/04/2026 12:13:10 - INFO - omnivoice.training.trainer - Running evaluation at step 30000...
|
| 24 |
+
05/04/2026 12:13:12 - INFO - omnivoice.training.trainer - Eval Loss: 4.0201
|
| 25 |
+
05/04/2026 12:13:12 - INFO - accelerate.accelerator - Saving current state to exp/omnivoice_vietnamese_4kh/checkpoint-30000
|
| 26 |
+
05/04/2026 12:13:21 - INFO - accelerate.checkpointing - Model weights saved in exp/omnivoice_vietnamese_4kh/checkpoint-30000/model.safetensors
|
| 27 |
+
05/04/2026 12:13:25 - INFO - accelerate.checkpointing - Optimizer state saved in exp/omnivoice_vietnamese_4kh/checkpoint-30000/optimizer.bin
|
| 28 |
+
05/04/2026 12:13:25 - INFO - accelerate.checkpointing - Scheduler state saved in exp/omnivoice_vietnamese_4kh/checkpoint-30000/scheduler.bin
|
| 29 |
+
05/04/2026 12:13:25 - INFO - accelerate.checkpointing - Random states saved in exp/omnivoice_vietnamese_4kh/checkpoint-30000/random_states_0.pkl
|
| 30 |
+
05/04/2026 12:13:27 - INFO - omnivoice.training.checkpoint - Saved checkpoint to exp/omnivoice_vietnamese_4kh/checkpoint-30000
|
| 31 |
+
05/04/2026 12:24:29 - INFO - omnivoice.training.trainer - Epoch 4 starting. Resetting dataloader...
|
| 32 |
+
05/04/2026 13:09:54 - INFO - omnivoice.training.trainer - Epoch 5 starting. Resetting dataloader...
|
| 33 |
+
05/04/2026 13:10:37 - INFO - omnivoice.training.trainer - Running evaluation at step 40000...
|
| 34 |
+
05/04/2026 13:10:38 - INFO - omnivoice.training.trainer - Eval Loss: 4.0533
|
| 35 |
+
05/04/2026 13:10:38 - INFO - accelerate.accelerator - Saving current state to exp/omnivoice_vietnamese_4kh/checkpoint-40000
|
| 36 |
+
05/04/2026 13:10:48 - INFO - accelerate.checkpointing - Model weights saved in exp/omnivoice_vietnamese_4kh/checkpoint-40000/model.safetensors
|
| 37 |
+
05/04/2026 13:10:52 - INFO - accelerate.checkpointing - Optimizer state saved in exp/omnivoice_vietnamese_4kh/checkpoint-40000/optimizer.bin
|
| 38 |
+
05/04/2026 13:10:52 - INFO - accelerate.checkpointing - Scheduler state saved in exp/omnivoice_vietnamese_4kh/checkpoint-40000/scheduler.bin
|
| 39 |
+
05/04/2026 13:10:52 - INFO - accelerate.checkpointing - Random states saved in exp/omnivoice_vietnamese_4kh/checkpoint-40000/random_states_0.pkl
|
| 40 |
+
05/04/2026 13:10:55 - INFO - omnivoice.training.checkpoint - Saved checkpoint to exp/omnivoice_vietnamese_4kh/checkpoint-40000
|
| 41 |
+
05/04/2026 13:55:42 - INFO - omnivoice.training.trainer - Epoch 6 starting. Resetting dataloader...
|
| 42 |
+
05/04/2026 14:08:00 - INFO - omnivoice.training.trainer - Running evaluation at step 50000...
|
| 43 |
+
05/04/2026 14:08:02 - INFO - omnivoice.training.trainer - Eval Loss: 4.0592
|
| 44 |
+
05/04/2026 14:08:02 - INFO - accelerate.accelerator - Saving current state to exp/omnivoice_vietnamese_4kh/checkpoint-50000
|
| 45 |
+
05/04/2026 14:08:12 - INFO - accelerate.checkpointing - Model weights saved in exp/omnivoice_vietnamese_4kh/checkpoint-50000/model.safetensors
|
| 46 |
+
05/04/2026 14:08:16 - INFO - accelerate.checkpointing - Optimizer state saved in exp/omnivoice_vietnamese_4kh/checkpoint-50000/optimizer.bin
|
| 47 |
+
05/04/2026 14:08:16 - INFO - accelerate.checkpointing - Scheduler state saved in exp/omnivoice_vietnamese_4kh/checkpoint-50000/scheduler.bin
|
| 48 |
+
05/04/2026 14:08:16 - INFO - accelerate.checkpointing - Random states saved in exp/omnivoice_vietnamese_4kh/checkpoint-50000/random_states_0.pkl
|
| 49 |
+
05/04/2026 14:08:18 - INFO - omnivoice.training.checkpoint - Saved checkpoint to exp/omnivoice_vietnamese_4kh/checkpoint-50000
|
| 50 |
+
05/04/2026 14:41:31 - INFO - omnivoice.training.trainer - Epoch 7 starting. Resetting dataloader...
|