Add files using upload-large-folder tool

README.md

@@ -0,0 +1,86 @@
+---
+license: apache-2.0
+base_model: openbmb/VoxCPM2
+tags:
+  - text-to-speech
+  - turkish
+  - voxcpm2
+  - merged-lora
+  - speech-synthesis
+  - trendyol-tts
+language:
+  - tr
+pipeline_tag: text-to-speech
+datasets:
+  - AlicanKiraz0/TR-Voice-TTS
+---
+
+# Trendyol-TTS
+
+Trendyol-TTS is a Turkish text-to-speech research model derived from the Kahya TTS `step_2000` checkpoint. It contains a full merged VoxCPM2 model: the Kahya Turkish LoRA adapter has been applied to the `openbmb/VoxCPM2` base weights so the repository can be loaded as a standalone model artifact.
+
+## Model Summary
+
+- **Model type:** Turkish text-to-speech / speech synthesis
+- **Base model:** `openbmb/VoxCPM2`
+- **Source checkpoint:** `AlicanKiraz0/Kahya-TTS-step_2000`
+- **Target repository:** `AlicanKiraz0/Trendyol-TTS`
+- **Fine-tuning method:** LoRA adapter merged into the base model weights
+- **Training dataset:** `AlicanKiraz0/TR-Voice-TTS`
+- **Primary language:** Turkish (`tr`)
+- **Recommended default checkpoint:** `step_0002000`
+
+## Contents
+
+This repository mirrors the merged `step_2000` model artifacts and includes:
+
+- `model.safetensors`: merged VoxCPM2 weights with the Turkish LoRA adapter applied.
+- `config.json`, tokenizer files, and `audiovae.pth`: runtime files inherited from `openbmb/VoxCPM2`.
+- `merge_manifest.json`: source model, adapter, checkpoint, and merge metadata.
+- `lora_adapter/`: original LoRA adapter files preserved for provenance and auditability.
+
+## Intended Use
+
+Trendyol-TTS is intended for Turkish TTS research, prototyping, and controlled internal evaluation. It is suitable for generating Turkish speech samples and comparing voice quality under fixed inference settings.
+
+This repository is **not** a production serving endpoint by itself. Before production use, validate latency, throughput, stability, monitoring, abuse prevention, and output quality on the intended serving stack.
+
+## Recommended Inference Settings
+
+The currently recommended clean default settings are:
+
+```text
+cfg_value = 2.0
+inference_timesteps = 16
+```
+
+A more expressive/natural setting to test is:
+
+```text
+cfg_value = 1.5
+inference_timesteps = 16
+```
+
+Avoid using `cfg_value=2.5` as a general production default. Prior audio proxy checks showed peaks getting too close to `0 dBFS` in some samples, even when clipping fraction was zero.
+
+## Evaluation Notes
+
+The source `step_2000` checkpoint was selected as the default based on human listening preference, a snappy response profile, and the absence of obvious production-level regressions in the available evaluation set. Later continuation checkpoints (`step_0002250` and `step_0002500`) remain internal archive/challenger artifacts and are not promoted as the default.
+
+Recent blind-evaluation tooling separates public listener packages from private mapping and resolved metrics. For release-grade evaluation, use public blind packages only during listening and generate post-unblind reports after mapping is opened.
+
+## Limitations
+
+- The model is a research artifact and should be validated before customer-facing deployment.
+- The training data is a private Turkish speech dataset; distribution and usage should respect the dataset owner policies.
+- Formal MOS, large-scale stress testing, ASR/CER semantic regression checks, and production serving validation are not included in this model repository.
+- Generated speech may still contain pronunciation, prosody, normalization, or artifact issues, especially for unusual names, abbreviations, numbers, and long-form text.
+- This model should not be used for impersonation, deception, unauthorized voice cloning, or any use that violates applicable law or platform policy.
+
+## Provenance
+
+This repository was copied from `AlicanKiraz0/Kahya-TTS-step_2000` and rebranded as `Trendyol-TTS`. The model weights are expected to be identical to the source merged `step_2000` model, except for the model card and repository metadata.
+
+## License
+
+The code/model artifact metadata follows the Apache-2.0 license declared by the upstream VoxCPM2 model where applicable. Users are responsible for checking the licenses and usage terms of `openbmb/VoxCPM2`, the dataset, and any downstream deployment environment.

audiovae.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:94b5d51e107e0507d4acc976cfdadb64edd6fd06d1f751dadbf2fd1594274bf1
+size 376951122

config.json

@@ -0,0 +1,67 @@
+{
+    "architecture": "voxcpm2",
+    "lm_config": {
+        "bos_token_id": 1,
+        "eos_token_id": 2,
+        "hidden_size": 2048,
+        "intermediate_size": 6144,
+        "max_position_embeddings": 32768,
+        "num_attention_heads": 16,
+        "num_hidden_layers": 28,
+        "num_key_value_heads": 2,
+        "rms_norm_eps": 1e-05,
+        "rope_theta": 10000,
+        "kv_channels": 128,
+        "rope_scaling": {
+            "type": "longrope", 
+            "long_factor": [0.9977997200264581, 1.014658295992452, 1.0349680404997148, 1.059429246056193, 1.0888815016813513, 1.1243301355211495, 1.166977103606075, 1.2182568066927284, 1.2798772354275727, 1.3538666751582975, 1.4426259039919596, 1.5489853358570191, 1.6762658237220625, 1.8283407612492941, 2.0096956085876183, 2.225478927469756, 2.481536379650452, 2.784415934557119, 3.1413289096347365, 3.560047844772632, 4.048719380066383, 4.615569542115128, 5.2684819496549835, 6.014438591970396, 6.858830049237097, 7.804668263503327, 8.851768731513417, 9.99600492938444, 11.228766118181639, 12.536757560834843, 13.902257701387796, 15.303885189125953, 16.717837610115794, 18.119465097853947, 19.484965238406907, 20.792956681060105, 22.02571786985731, 23.16995406772833, 24.217054535738416, 25.16289275000465, 26.007284207271347, 26.753240849586767, 27.40615325712662, 27.973003419175363, 28.461674954469114, 28.880393889607006, 29.237306864684626, 29.540186419591297, 29.79624387177199, 30.01202719065413, 30.193382037992453, 30.34545697551969, 30.47273746338473, 30.579096895249787, 30.66785612408345, 30.741845563814174, 30.80346599254902, 30.85474569563567, 30.897392663720595, 30.932841297560394, 30.962293553185553, 30.986754758742034, 31.007064503249293, 31.02392307921529],
+            "short_factor": [0.9977997200264581, 1.014658295992452, 1.0349680404997148, 1.059429246056193, 1.0888815016813513, 1.1243301355211495, 1.166977103606075, 1.2182568066927284, 1.2798772354275727, 1.3538666751582975, 1.4426259039919596, 1.5489853358570191, 1.6762658237220625, 1.8283407612492941, 2.0096956085876183, 2.225478927469756, 2.481536379650452, 2.784415934557119, 3.1413289096347365, 3.560047844772632, 4.048719380066383, 4.615569542115128, 5.2684819496549835, 6.014438591970396, 6.858830049237097, 7.804668263503327, 8.851768731513417, 9.99600492938444, 11.228766118181639, 12.536757560834843, 13.902257701387796, 15.303885189125953, 16.717837610115794, 18.119465097853947, 19.484965238406907, 20.792956681060105, 22.02571786985731, 23.16995406772833, 24.217054535738416, 25.16289275000465, 26.007284207271347, 26.753240849586767, 27.40615325712662, 27.973003419175363, 28.461674954469114, 28.880393889607006, 29.237306864684626, 29.540186419591297, 29.79624387177199, 30.01202719065413, 30.193382037992453, 30.34545697551969, 30.47273746338473, 30.579096895249787, 30.66785612408345, 30.741845563814174, 30.80346599254902, 30.85474569563567, 30.897392663720595, 30.932841297560394, 30.962293553185553, 30.986754758742034, 31.007064503249293, 31.02392307921529],
+            "original_max_position_embeddings": 32768
+        },    
+        "vocab_size": 73448,
+        "use_mup": false,
+        "scale_emb": 12,
+        "dim_model_base": 256,
+        "scale_depth": 1.4
+    },
+    "patch_size": 4,
+    "feat_dim": 64,
+    "scalar_quantization_latent_dim": 512,
+    "scalar_quantization_scale": 9,
+    "residual_lm_num_layers": 8,
+    "residual_lm_no_rope": true,
+    "encoder_config": {
+        "hidden_dim": 1024,
+        "ffn_dim": 4096,
+        "num_heads": 16,
+        "num_layers": 12,
+        "kv_channels": 128
+    },
+    "dit_config": {
+        "hidden_dim": 1024,
+        "ffn_dim": 4096,
+        "num_heads": 16,
+        "num_layers": 12,
+        "kv_channels": 128,
+        "mean_mode": false,
+        "cfm_config": {
+            "sigma_min": 1e-06,
+            "solver": "euler",
+            "t_scheduler": "log-norm",
+            "inference_cfg_rate": 2.0
+        }
+    },
+    "audio_vae_config": {
+        "encoder_dim": 128,
+        "encoder_rates": [2, 5, 8, 8],
+        "latent_dim": 64,
+        "decoder_dim": 2048,
+        "decoder_rates": [8, 6, 5, 2, 2, 2],
+        "sr_bin_boundaries": [20000, 30000, 40000],
+        "sample_rate": 16000,
+        "out_sample_rate": 48000
+    },
+    "max_length": 8192,
+    "device": "cuda",
+    "dtype": "bfloat16"
+}

lora_adapter/configs/voxcpm_finetune_lora.yaml

@@ -0,0 +1,34 @@
+pretrained_path: /workspace/models/VoxCPM2/
+train_manifest: /workspace/data/voxcpm_manifests/train.jsonl
+val_manifest: /workspace/data/voxcpm_manifests/val.jsonl
+
+sample_rate: 16000
+out_sample_rate: 48000
+batch_size: 2
+grad_accum_steps: 8
+num_workers: 2
+num_iters: 2000
+log_interval: 10
+valid_interval: 250
+save_interval: 250
+
+learning_rate: 0.0001
+weight_decay: 0.01
+warmup_steps: 200
+max_steps: 2000
+max_batch_tokens: 8192
+
+save_path: /workspace/checkpoints/lora
+tensorboard: /workspace/logs/lora
+
+lambdas:
+  loss/diff: 1.0
+  loss/stop: 1.0
+
+lora:
+  enable_lm: true
+  enable_dit: true
+  enable_proj: false
+  r: 64
+  alpha: 64
+  dropout: 0.0

lora_adapter/lora_config.json

@@ -0,0 +1,29 @@
+{
+  "base_model": "/workspace/models/VoxCPM2/",
+  "lora_config": {
+    "enable_lm": true,
+    "enable_dit": true,
+    "enable_proj": false,
+    "r": 64,
+    "alpha": 64,
+    "dropout": 0.0,
+    "target_modules_lm": [
+      "q_proj",
+      "v_proj",
+      "k_proj",
+      "o_proj"
+    ],
+    "target_modules_dit": [
+      "q_proj",
+      "v_proj",
+      "k_proj",
+      "o_proj"
+    ],
+    "target_proj_modules": [
+      "enc_to_lm_proj",
+      "lm_to_dit_proj",
+      "res_to_dit_proj",
+      "fusion_concat_proj"
+    ]
+  }
+}

lora_adapter/lora_weights.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d3cf875aa83035a802234fe6b750e03c3b17d956e1ae6ce36c6b0ffaadd7afa9
+size 144749240

lora_adapter/training_state.json

@@ -0,0 +1 @@
+{"step": 2000}

merge_manifest.json

@@ -0,0 +1,40 @@
+{
+  "base_model": "openbmb/VoxCPM2",
+  "adapter_source": "AlicanKiraz0/Kahya-TTS-Turkish",
+  "checkpoint_step": 2000,
+  "training_state": {
+    "step": 2000
+  },
+  "lora_config": {
+    "base_model": "/workspace/models/VoxCPM2/",
+    "lora_config": {
+      "enable_lm": true,
+      "enable_dit": true,
+      "enable_proj": false,
+      "r": 64,
+      "alpha": 64,
+      "dropout": 0.0,
+      "target_modules_lm": [
+        "q_proj",
+        "v_proj",
+        "k_proj",
+        "o_proj"
+      ],
+      "target_modules_dit": [
+        "q_proj",
+        "v_proj",
+        "k_proj",
+        "o_proj"
+      ],
+      "target_proj_modules": [
+        "enc_to_lm_proj",
+        "lm_to_dit_proj",
+        "res_to_dit_proj",
+        "fusion_concat_proj"
+      ]
+    }
+  },
+  "merge_scale": 1.0,
+  "merged_target_weights": 192,
+  "base_snapshot": "bffb3df5a29440629464e5e839f4d214c8714c3d"
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:70073e6d60adbceceea60ba3eb027fb5c57688cf40553f6e221a5546e7c87f9a
+size 4580080752

special_tokens_map.json

@@ -0,0 +1,81 @@
+{
+  "additional_special_tokens": [
+    {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<|tool_call|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<|execute_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<|execute_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    },
+    {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false
+    }
+  ],
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenization_voxcpm2.py

@@ -0,0 +1,72 @@
+"""Custom tokenizer for VoxCPM2 that splits multi-character Chinese tokens.
+
+VoxCPM2 was trained with ``mask_multichar_chinese_tokens`` which splits
+multi-character Chinese tokens (e.g. "你好" -> ["你", "好"]) into individual
+character IDs before embedding.  The base LlamaTokenizerFast produces
+multi-character Chinese tokens that the model has never seen during training,
+yielding garbled Chinese audio output in downstream inference frameworks.
+
+This module provides ``VoxCPM2Tokenizer`` which transparently applies the
+character splitting inside ``encode()`` and ``__call__()``, so any downstream
+consumer (vLLM, vLLM-Omni, Nano-vLLM, etc.) gets correct single-character
+IDs without code changes.
+"""
+
+from transformers import LlamaTokenizerFast
+
+
+class VoxCPM2Tokenizer(LlamaTokenizerFast):
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        self._split_map = self._build_split_map()
+
+    def _build_split_map(self) -> dict[int, list[int]]:
+        vocab = self.get_vocab()
+        split_map: dict[int, list[int]] = {}
+        for token, tid in vocab.items():
+            clean = token.replace("\u2581", "")
+            if len(clean) >= 2 and all(self._is_cjk(c) for c in clean):
+                char_ids = self.convert_tokens_to_ids(list(clean))
+                if all(c != self.unk_token_id for c in char_ids):
+                    split_map[tid] = char_ids
+        return split_map
+
+    @staticmethod
+    def _is_cjk(c: str) -> bool:
+        return (
+            "\u4e00" <= c <= "\u9fff"
+            or "\u3400" <= c <= "\u4dbf"
+            or "\uf900" <= c <= "\ufaff"
+            or "\U00020000" <= c <= "\U0002a6df"
+        )
+
+    def _expand_ids(self, ids: list[int]) -> list[int]:
+        result: list[int] = []
+        for tid in ids:
+            expansion = self._split_map.get(tid)
+            if expansion is not None:
+                result.extend(expansion)
+            else:
+                result.append(tid)
+        return result
+
+    def encode(self, text, *args, **kwargs):
+        ids = super().encode(text, *args, **kwargs)
+        return self._expand_ids(ids)
+
+    def __call__(self, text, *args, **kwargs):
+        result = super().__call__(text, *args, **kwargs)
+        if hasattr(result, "input_ids"):
+            ids = result["input_ids"]
+            if isinstance(ids, list) and ids and isinstance(ids[0], list):
+                result["input_ids"] = [self._expand_ids(x) for x in ids]
+                if "attention_mask" in result:
+                    result["attention_mask"] = [
+                        [1] * len(x) for x in result["input_ids"]
+                    ]
+            elif isinstance(ids, list):
+                result["input_ids"] = self._expand_ids(ids)
+                if "attention_mask" in result:
+                    result["attention_mask"] = [1] * len(result["input_ids"])
+        return result

tokenizer_config.json

@@ -0,0 +1,218 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "101": {
+      "content": "<|audio_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "102": {
+      "content": "<|audio_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "103": {
+      "content": "<|audio_prompt_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "104": {
+      "content": "<|audio_prompt_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "105": {
+      "content": "<|background|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "106": {
+      "content": "<|/background|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "107": {
+      "content": "<|characters|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "108": {
+      "content": "<|/characters|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "109": {
+      "content": "<|speaker_id|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "110": {
+      "content": "<|/speaker_id|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "111": {
+      "content": "<|span|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "112": {
+      "content": "<|/span|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "73440": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "73441": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "73442": {
+      "content": "<|tool_call|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "73443": {
+      "content": "<|execute_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "73444": {
+      "content": "<|execute_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "73445": {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "73446": {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "73447": {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_end|>",
+    "<|im_start|>",
+    "<|tool_call|>",
+    "<|execute_start|>",
+    "<|execute_end|>",
+    "<|fim_prefix|>",
+    "<|fim_middle|>",
+    "<|fim_suffix|>"
+  ],
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "legacy": true,
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": null,
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "VoxCPM2Tokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false,
+  "chat_template": "{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
+  "auto_map": {
+    "AutoTokenizer": [
+      "tokenization_voxcpm2.VoxCPM2Tokenizer",
+      null
+    ]
+  }
+}