Upload folder using huggingface_hub

F5TTS_v1_Base_Open_Bible_Dholuo.yaml

@@ -0,0 +1,55 @@
+hydra:
+  run:
+    dir: ckpts/${model.name}_${model.mel_spec.mel_spec_type}_${model.tokenizer}_${datasets.name}/${now:%Y-%m-%d}/${now:%H-%M-%S}
+
+datasets:
+  name: open-bible-dholuo  # dataset name
+  batch_size_per_gpu: 28000  # 8 GPUs, 8 * 38400 = 307200
+  batch_size_type: frame  # frame | sample
+  max_samples: 32  # max sequences per batch if use frame-wise batch_size. we set 32 for small models, 64 for base models
+  num_workers: 4
+
+optim:
+  epochs: 549
+  learning_rate: 7.5e-5
+  num_warmup_updates: 20000  # warmup updates
+  grad_accumulation_steps: 1  # note: updates = steps / grad_accumulation_steps
+  max_grad_norm: 1.0  # gradient clipping
+  bnb_optimizer: False  # use bnb 8bit AdamW optimizer or not
+
+model:
+  name: F5TTS_v1_Base  # model name
+  tokenizer: custom  # tokenizer type
+  tokenizer_path: data/open-bible-dholuo_custom/vocab.txt  # if 'custom' tokenizer, define the path want to use (should be vocab.txt)
+  backbone: DiT
+  arch:
+    dim: 1024
+    depth: 22
+    heads: 16
+    ff_mult: 2
+    text_dim: 512
+    text_mask_padding: True
+    qk_norm: null  # null | rms_norm
+    conv_layers: 4
+    pe_attn_head: null
+    attn_backend: torch  # torch | flash_attn
+    attn_mask_enabled: False
+    checkpoint_activations: False  # recompute activations and save memory for extra compute
+  mel_spec:
+    target_sample_rate: 24000
+    n_mel_channels: 100
+    hop_length: 256
+    win_length: 1024
+    n_fft: 1024
+    mel_spec_type: vocos  # vocos | bigvgan
+  vocoder:
+    is_local: False  # use local offline ckpt or not
+    local_path: null  # local vocoder path
+
+ckpts:
+  logger: wandb #| tensorboard | null
+  log_samples: True  # infer random sample per save checkpoint. wip, normal to fail with extra long samples
+  save_per_updates: 10000  # save checkpoint per updates
+  keep_last_n_checkpoints: 5  # -1 to keep all, 0 to not save intermediate, > 0 to keep last N checkpoints
+  last_per_updates: 5000  # save last checkpoint per updates
+  save_dir: ckpts/${model.name}_${model.mel_spec.mel_spec_type}_${model.tokenizer}_${datasets.name}

model_last.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c53e18a68c0ad3138513e25c74717a3752e0803bd05df5452c9259be5cf6fa2
+size 5373999779

vocab.txt

@@ -0,0 +1,80 @@
+ 
+!
+"
+'
+(
+)
+*
++
+,
+-
+.
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+?
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+R
+S
+T
+U
+V
+W
+Y
+Z
+[
+\
+]
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+ʼ
+ט
+י
+ר