sophon-realtime 1.0.0__tar.gz

sophon_realtime-1.0.0/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

sophon_realtime-1.0.0/NOTICE

@@ -0,0 +1,9 @@
+Copyright 2026 Adventists.ai.
+
+Sophon-Realtime-32B-Tag bundles model weights derived from the following works, each licensed under the Apache License 2.0:
+- thinker/: derived from Qwen3-VL-32B-Instruct by the Qwen team, Alibaba (https://huggingface.co/Qwen/Qwen3-VL-32B-Instruct),
+  via DuplexJev-32B-Para (Adventists.ai) with a merged tag-writing adapter; quantized to FP8. Its audio encoder is derived from
+  Qwen3-ASR (Qwen team, Alibaba, Apache License 2.0).
+- talker/: tag-tts (Adventists.ai), derived from Fun-CosyVoice3-0.5B by the FunAudioLLM team, Alibaba
+  (https://github.com/FunAudioLLM/CosyVoice); the language model was fine-tuned to perform sound tags, the flow model was
+  distilled to 4 steps, and the HiFT vocoder is redistributed unchanged.

sophon_realtime-1.0.0/PKG-INFO

@@ -0,0 +1,38 @@
+Metadata-Version: 2.4
+Name: sophon-realtime
+Version: 1.0.0
+Summary: Sophon-Realtime v1: open full-duplex voice model that hears your tone and speaks with one (Tag build, one GPU, vLLM-Omni)
+Author: Adventists AI
+License-Expression: Apache-2.0
+Project-URL: Homepage, https://github.com/adventists-ai/sophon-realtime
+Project-URL: ModelScope, https://www.modelscope.cn/models/AdventistsAI/Sophon-Realtime-v1
+Project-URL: Hugging Face, https://huggingface.co/adventists-ai/Sophon-Realtime-32B-Tag
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+License-File: NOTICE
+Requires-Dist: vllm==0.30.0
+Requires-Dist: vllm-omni==0.30.0
+Requires-Dist: duplexjev-vllm==0.3.0
+Requires-Dist: modelscope
+Provides-Extra: trt
+Requires-Dist: tensorrt-cu13==10.13.3.9; extra == "trt"
+Dynamic: license-file
+
+# sophon-realtime
+
+**Sophon-Realtime v1**: an open full-duplex voice model that hears your tone and speaks with one. This package installs the recommended **Tag** build as a vLLM-Omni plugin, plus a small command line to download and serve it on one 80 GB GPU.
+
+```bash
+pip install sophon-realtime          # Python 3.12, Linux, CUDA; installs vLLM 0.30 + vLLM-Omni 0.30
+sophon-realtime serve                # downloads Sophon-Realtime-32B-Tag (~42 GB) on first run, then serves on port 8000
+```
+
+- Full duplex: `ws://HOST:8000/v1/realtime` (turn-taking, barge-in, back-channels; OpenAI-Realtime-style)
+- `POST /v1/chat/completions` (audio in, text + audio out)
+- Options: `--port`, `--gpu`, `--model DIR`, `--source modelscope|hf`; `sophon-realtime download` only downloads.
+- Optional TensorRT for faster speech: `pip install "sophon-realtime[trt]"`.
+
+VoiceBench 82.8 and Big Bench Audio 81.8% (#1 among open full-duplex models), #1 in paralinguistic understanding, NVV-SuperBench 0.86 (zh) / 0.80 (en). Client examples, the Flash build, Live mode (OpenAI GPT-Live compatible) and fine-tuning: https://github.com/adventists-ai/sophon-realtime
+
+Apache-2.0. Third-party notices: NOTICE.

sophon_realtime-1.0.0/README.md

@@ -0,0 +1,17 @@
+# sophon-realtime
+
+**Sophon-Realtime v1**: an open full-duplex voice model that hears your tone and speaks with one. This package installs the recommended **Tag** build as a vLLM-Omni plugin, plus a small command line to download and serve it on one 80 GB GPU.
+
+```bash
+pip install sophon-realtime          # Python 3.12, Linux, CUDA; installs vLLM 0.30 + vLLM-Omni 0.30
+sophon-realtime serve                # downloads Sophon-Realtime-32B-Tag (~42 GB) on first run, then serves on port 8000
+```
+
+- Full duplex: `ws://HOST:8000/v1/realtime` (turn-taking, barge-in, back-channels; OpenAI-Realtime-style)
+- `POST /v1/chat/completions` (audio in, text + audio out)
+- Options: `--port`, `--gpu`, `--model DIR`, `--source modelscope|hf`; `sophon-realtime download` only downloads.
+- Optional TensorRT for faster speech: `pip install "sophon-realtime[trt]"`.
+
+VoiceBench 82.8 and Big Bench Audio 81.8% (#1 among open full-duplex models), #1 in paralinguistic understanding, NVV-SuperBench 0.86 (zh) / 0.80 (en). Client examples, the Flash build, Live mode (OpenAI GPT-Live compatible) and fine-tuning: https://github.com/adventists-ai/sophon-realtime
+
+Apache-2.0. Third-party notices: NOTICE.

sophon_realtime-1.0.0/pyproject.toml

@@ -0,0 +1,36 @@
+[build-system]
+requires = ["setuptools>=61"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "sophon-realtime"
+version = "1.0.0"
+description = "Sophon-Realtime v1: open full-duplex voice model that hears your tone and speaks with one (Tag build, one GPU, vLLM-Omni)"
+readme = "README.md"
+requires-python = ">=3.10"
+license = "Apache-2.0"
+authors = [{name = "Adventists AI"}]
+dependencies = ["vllm==0.30.0", "vllm-omni==0.30.0", "duplexjev-vllm==0.3.0", "modelscope"]
+
+[project.optional-dependencies]
+trt = ["tensorrt-cu13==10.13.3.9"]
+
+[project.urls]
+Homepage = "https://github.com/adventists-ai/sophon-realtime"
+ModelScope = "https://www.modelscope.cn/models/AdventistsAI/Sophon-Realtime-v1"
+"Hugging Face" = "https://huggingface.co/adventists-ai/Sophon-Realtime-32B-Tag"
+
+[project.scripts]
+sophon-realtime = "sophon_realtime.cli:main"
+
+[project.entry-points."vllm_omni.general_plugins"]
+sophon_realtime = "sophon_realtime:register"
+
+[project.entry-points."vllm.general_plugins"]
+sophon_realtime = "sophon_realtime:register"
+
+[tool.setuptools.packages.find]
+include = ["sophon_realtime*"]
+
+[tool.setuptools.package-data]
+sophon_realtime = ["deploy/*.yaml", "deploy/*.onnx", "deploy/*.json"]

sophon_realtime-1.0.0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

sophon_realtime-1.0.0/sophon_realtime/__init__.py

@@ -0,0 +1,53 @@
+"""Sophon-Realtime-32B-Tag: DuplexJev (listen + write, with sound tags) -> tag-tts talker (CosyVoice3 LM, text streamed in
+bistream style while it is written) -> CosyVoice3 flow + HiFT, served as one vLLM-Omni model.
+
+vllm serve <Sophon-Realtime-32B-Tag dir> --omni --deploy-config <sophon_realtime_tag_1gpu.yaml>
+"""
+__version__ = "1.0.0"
+ARCH = "SophonRealtimeForConditionalGeneration"
+TALKER_ARCH = "SophonTagTalker"
+_done = False
+
+
+def register():
+    global _done
+    if _done:
+        return
+    _done = True
+    from transformers import AutoConfig
+    from .config import SophonRealtimeConfig
+    try:
+        AutoConfig.register(SophonRealtimeConfig.model_type, SophonRealtimeConfig)
+    except ValueError:
+        pass
+    try:
+        from vllm_omni.transformers_utils.configs.cosyvoice3 import CosyVoice3Config
+        AutoConfig.register("cosyvoice3", CosyVoice3Config)
+    except Exception:  # noqa: BLE001  already registered by vllm-omni
+        pass
+    try:  # the talker re-emits the voice prompt (embed.*): keep the latest instead of concatenating per step
+        from vllm_omni.outputs.output_modality import TensorAccumulationStrategy, register_key_accumulation_strategy
+        for k in ("embed.speech_token", "embed.speech_feat", "embed.embedding", "embed.speech_token_len"):
+            register_key_accumulation_strategy(k, TensorAccumulationStrategy.REPLACE)
+    except Exception:  # noqa: BLE001
+        pass
+    try:
+        import duplexjev_vllm
+        duplexjev_vllm.register()
+    except Exception:  # noqa: BLE001
+        pass
+    from vllm import ModelRegistry
+    if TALKER_ARCH not in ModelRegistry.get_supported_archs():
+        ModelRegistry.register_model(TALKER_ARCH, "sophon_realtime.talker_tag:SophonTagTalker")
+    try:
+        from vllm.model_executor.models.registry import _LazyRegisteredModel
+        from vllm_omni.model_executor.models.registry import OmniModelRegistry
+        OmniModelRegistry.models.setdefault(TALKER_ARCH, _LazyRegisteredModel(module_name="sophon_realtime.talker_tag", class_name="SophonTagTalker"))
+        OmniModelRegistry.models.setdefault("DuplexJevForConditionalGeneration", _LazyRegisteredModel(module_name="duplexjev_vllm.model", class_name="DuplexJevForConditionalGeneration"))
+        OmniModelRegistry.models.setdefault("DuplexJevVisionForConditionalGeneration", _LazyRegisteredModel(module_name="duplexjev_vllm.vision_model", class_name="DuplexJevVisionForConditionalGeneration"))
+        from vllm_omni.config.pipeline_registry import OMNI_PIPELINES, register_pipeline
+        from .pipeline import SOPHON_REALTIME_PIPELINE
+        if SOPHON_REALTIME_PIPELINE.model_type not in OMNI_PIPELINES:
+            register_pipeline(SOPHON_REALTIME_PIPELINE)
+    except ImportError:
+        pass

sophon_realtime-1.0.0/sophon_realtime/bridge.py

@@ -0,0 +1,79 @@
+"""Client tool bridge: lets a gateway outside the serving process see the model's tool calls and hand back results.
+
+Engine-agnostic contract (the same two calls are implemented by every engine adapter):
+  GET  /bridge/<key>/events?wait=<s>   -> {"events": [{"type": "tool_call", "call_id", "name", "arguments"}]}  (long poll)
+  POST /bridge/<key>/result            <- {"call_id": str|null, "output": str, "speak": bool}
+<key> is the session's extra_body.bridge_key. Enabled when SOPHON_BRIDGE_PORT is set; binds 127.0.0.1 only."""
+from __future__ import annotations
+
+import json
+import os
+import threading
+import time
+from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
+from urllib.parse import parse_qs, urlparse
+
+PORT = int(os.environ.get("SOPHON_BRIDGE_PORT", "0") or 0)
+
+
+class Bridge:
+    def __init__(self, on_result):
+        self.on_result = on_result                 # (key, call_id, output, speak) -> bool
+        self.q: dict[str, list] = {}
+        self.cv = threading.Condition()
+        self.seen: dict[str, float] = {}
+        self._srv = None
+
+    def start(self):
+        if not PORT or self._srv is not None:
+            return
+        br = self
+
+        class H(BaseHTTPRequestHandler):
+            def log_message(self, *a):
+                pass
+
+            def _send(self, code, obj):
+                b = json.dumps(obj, ensure_ascii=False).encode()
+                try:
+                    self.send_response(code); self.send_header("Content-Type", "application/json"); self.send_header("Content-Length", str(len(b))); self.end_headers(); self.wfile.write(b)
+                except (BrokenPipeError, ConnectionResetError):
+                    pass                                  # the gateway closed the long poll (session ended)
+
+            def do_GET(self):
+                u = urlparse(self.path); p = u.path.strip("/").split("/")
+                if len(p) == 3 and p[0] == "bridge" and p[2] == "events":
+                    wait = min(float(parse_qs(u.query).get("wait", ["20"])[0]), 60.0)
+                    return self._send(200, {"events": br.pop(p[1], wait)})
+                self._send(404, {"error": "not found"})
+
+            def do_POST(self):
+                p = urlparse(self.path).path.strip("/").split("/")
+                n = int(self.headers.get("Content-Length") or 0)
+                try:
+                    j = json.loads(self.rfile.read(n) or b"{}")
+                except Exception:  # noqa: BLE001
+                    return self._send(400, {"error": "bad json"})
+                if len(p) == 3 and p[0] == "bridge" and p[2] == "result":
+                    ok = br.on_result(p[1], j.get("call_id"), str(j.get("output", "")), bool(j.get("speak", True)))
+                    return self._send(200 if ok else 404, {"ok": ok})
+                self._send(404, {"error": "not found"})
+
+        self._srv = ThreadingHTTPServer(("127.0.0.1", PORT), H)
+        threading.Thread(target=self._srv.serve_forever, daemon=True, name="sophon-bridge").start()
+
+    def push(self, key, ev):
+        with self.cv:
+            self.q.setdefault(key, []).append(ev); self.cv.notify_all()
+
+    def pop(self, key, wait):
+        end = time.monotonic() + wait
+        with self.cv:
+            self.seen[key] = time.monotonic()
+            while not self.q.get(key):
+                left = end - time.monotonic()
+                if left <= 0:
+                    return []
+                self.cv.wait(left)
+            out, self.q[key] = self.q[key], []
+            return out

sophon_realtime-1.0.0/sophon_realtime/cli.py

@@ -0,0 +1,102 @@
+"""sophon-realtime command line: download the Tag weights and serve them with vLLM-Omni on one GPU.
+
+  sophon-realtime download [--dir DIR] [--source modelscope|hf]
+  sophon-realtime serve [--model DIR] [--port 8000] [--gpu 0]
+
+`serve` downloads Sophon-Realtime-32B-Tag first if --model does not hold it.
+Endpoints: ws://HOST:PORT/v1/realtime (full duplex) and POST /v1/chat/completions (audio in -> text + audio out).
+Live mode (OpenAI GPT-Live compatible) and examples: https://github.com/adventists-ai/sophon-realtime
+"""
+import argparse, os, shutil, subprocess, sys
+
+REPO = {"modelscope": "AdventistsAI/Sophon-Realtime-32B-Tag", "hf": "adventists-ai/Sophon-Realtime-32B-Tag"}
+DEFAULT_DIR = "Sophon-Realtime-32B-Tag"
+HERE = os.path.dirname(os.path.abspath(__file__))
+
+
+def _is_tag(d):
+    return os.path.isfile(os.path.join(d, "config.json")) and os.path.isdir(os.path.join(d, "thinker")) and os.path.isfile(
+        os.path.join(d, "talker", "flow.decoder.estimator.fp32.onnx"))
+
+
+def _source(s):
+    if s != "auto":
+        return s
+    try:
+        import modelscope  # noqa: F401
+        return "modelscope"
+    except ImportError:
+        return "hf"
+
+
+def download(dirname, source="auto"):
+    src = _source(source)
+    print(f"downloading {REPO[src]} ({src}) -> {dirname}  (about 42 GB)", flush=True)
+    if src == "modelscope":
+        from modelscope import snapshot_download
+        snapshot_download(REPO[src], local_dir=dirname)
+    else:
+        from huggingface_hub import snapshot_download
+        snapshot_download(REPO[src], local_dir=dirname)
+    return dirname
+
+
+def serve(a):
+    model = os.path.abspath(a.model)
+    if not _is_tag(model):
+        if os.path.isdir(model) and os.listdir(model):
+            print(f"{a.model} does not hold Sophon-Realtime-32B-Tag weights; downloading them to ./{DEFAULT_DIR}", flush=True)
+            model = os.path.abspath(DEFAULT_DIR)
+        if not _is_tag(model):
+            download(model, a.source)
+    os.makedirs(a.log_dir, exist_ok=True)
+    cfg_src = os.path.join(HERE, "deploy", "sophon_realtime_tag_1gpu.yaml")
+    cfg = os.path.join(os.path.abspath(a.log_dir), f"deploy_{a.port}.yaml")
+    with open(cfg_src) as f:
+        txt = f.read().replace("__HERE__/config/silero_vad.onnx", os.path.join(HERE, "deploy", "silero_vad.onnx"))
+    with open(cfg, "w") as f:
+        f.write(txt)
+    try:
+        import tensorrt  # noqa: F401
+        trt = "1"
+    except ImportError:
+        trt = "0"
+    env = dict(os.environ)
+    env.setdefault("CUDA_VISIBLE_DEVICES", str(a.gpu))
+    env.update({k: env.get(k, v) for k, v in {
+        "SOPHON_MODEL": model, "SOPHON_PORT": str(a.port), "SOPHON_BRIDGE_PORT": str(a.port + 700),
+        "VLLM_DISABLE_COMPILE_CACHE": "1", "COSYVOICE3_TRT": trt,
+        "SOPHON_FLOW_STEPS": "4", "SOPHON_FLOW_CFG": "0", "COSYVOICE3_BATCH_FLOW": "1", "SOPHON_FLOW_MAXB": "16",
+        "SOPHON_FLOW_ONNX": os.path.join(model, "talker", "flow.decoder.estimator.fp32.onnx")}.items()})
+    vllm = shutil.which("vllm") or "vllm"
+    cmd = [vllm, "serve", model, "--omni", "--host", a.host, "--port", str(a.port), "--trust-remote-code",
+           "--init-timeout", "3000", "--stage-init-timeout", "3000", "--deploy-config", cfg]
+    print("first start builds kernels / TensorRT engine: 5-20 min; later 5-8 min", flush=True)
+    print(f"when ready: ws://HOST:{a.port}/v1/realtime   http://HOST:{a.port}/v1/chat/completions", flush=True)
+    os.execvpe(cmd[0], cmd, env)
+
+
+def main(argv=None):
+    ap = argparse.ArgumentParser(prog="sophon-realtime", description="Sophon-Realtime v1 (Tag build) on one 80 GB GPU with vLLM-Omni.")
+    sub = ap.add_subparsers(dest="cmd")
+    d = sub.add_parser("download", help="download Sophon-Realtime-32B-Tag")
+    d.add_argument("--dir", default=DEFAULT_DIR); d.add_argument("--source", default="auto", choices=["auto", "modelscope", "hf"])
+    s = sub.add_parser("serve", help="serve the Tag build (downloads the weights if needed)")
+    s.add_argument("--model", default=DEFAULT_DIR, help=f"weights directory (default ./{DEFAULT_DIR})")
+    s.add_argument("--host", default="0.0.0.0"); s.add_argument("--port", type=int, default=8000)
+    s.add_argument("--gpu", default="0", help="GPU index (ignored if CUDA_VISIBLE_DEVICES is set)")
+    s.add_argument("--source", default="auto", choices=["auto", "modelscope", "hf"])
+    s.add_argument("--log-dir", default="logs")
+    a = ap.parse_args(argv)
+    if a.cmd == "download":
+        download(a.dir, a.source)
+    elif a.cmd == "serve":
+        serve(a)
+    else:
+        ap.print_help()
+        return 1
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())

sophon_realtime-1.0.0/sophon_realtime/config.py

@@ -0,0 +1,11 @@
+from transformers import PretrainedConfig
+
+
+class SophonRealtimeConfig(PretrainedConfig):
+    """Root config: only describes the layout (thinker/, talker/, voices/ subdirs); each stage's weights live in its own subdir."""
+    model_type = "sophon_realtime"
+
+    def __init__(self, thinker_subdir="thinker", talker_subdir="talker", voices_subdir="voices", default_voice="chelsie", default_language="zh", **kw):
+        self.thinker_subdir = thinker_subdir; self.talker_subdir = talker_subdir; self.voices_subdir = voices_subdir
+        self.default_voice = default_voice; self.default_language = default_language
+        super().__init__(**kw)

sophon_realtime-1.0.0/sophon_realtime/deploy/sophon_realtime_tag_1gpu.yaml

@@ -0,0 +1,79 @@
+# Sophon-Realtime-32B-Tag, ONE GPU (H100/A100 80 GB): thinker = offline FP8 (tag-writing merged), tag-tts talker + flow/HiFT on the same card.
+# vllm serve <Sophon-Realtime-32B-Tag> --omni --deploy-config sophon_realtime_tag_1gpu.yaml
+async_chunk: true
+session_mode: duplex
+duplex_session:
+  max_sessions: 32
+  idle_ttl_s: 60
+  server_vad_model_path: __HERE__/config/silero_vad.onnx
+
+connectors:
+  connector_of_shared_memory:
+    name: SharedMemoryConnector
+    extra:
+      codec_streaming: true
+      connector_get_sleep_s: 0.002
+      connector_get_max_wait_first_chunk: 3000
+      connector_get_max_wait: 300
+      codec_chunk_frames: 10          # first flow chunk = 10 speech tokens (hop 10, voice prompts are trimmed to a multiple of 10)
+      codec_left_context_frames: 25
+      codec_vocab_size: 6561
+
+stages:
+  - stage_id: 0
+    max_num_seqs: 32
+    gpu_memory_utilization: 0.62
+    trust_remote_code: true
+    enable_prefix_caching: true
+    max_model_len: 8192
+    mm_processor_cache_gb: 0
+    devices: "0"
+    limit_mm_per_prompt:
+      audio: 20
+      image: 2
+    output_connectors:
+      to_stage_1: connector_of_shared_memory
+    default_sampling_params:
+      temperature: 0.7
+      top_p: 0.9
+      max_tokens: 400
+
+  - stage_id: 1
+    max_num_seqs: 32
+    max_num_batched_tokens: 4096
+    gpu_memory_utilization: 0.08
+    enforce_eager: false
+    trust_remote_code: true
+    enable_prefix_caching: false
+    async_scheduling: false
+    max_model_len: 4096
+    devices: "0"
+    hf_overrides:
+      async_chunk_prewarm_prompt_len: 8     # [sos] + "You are a helpful assistant.<|endofprompt|>" (7 tokens)
+    input_connectors:
+      from_stage_0: connector_of_shared_memory
+    output_connectors:
+      to_stage_2: connector_of_shared_memory
+    default_sampling_params:
+      max_tokens: 2048
+      top_k: 25
+      top_p: 0.8
+    disable_hybrid_kv_cache_manager: true
+    mm_processor_cache_gb: 0
+    skip_mm_profiling: true
+
+  - stage_id: 2
+    max_num_batched_tokens: 32768
+    max_num_seqs: 32
+    gpu_memory_utilization: 0.08
+    enforce_eager: true
+    trust_remote_code: true
+    enable_prefix_caching: false
+    max_model_len: 32768
+    devices: "0"
+    input_connectors:
+      from_stage_1: connector_of_shared_memory
+    default_sampling_params:
+      max_tokens: 2048
+    disable_hybrid_kv_cache_manager: true
+    skip_mm_profiling: true

sophon_realtime-1.0.0/sophon_realtime/deploy/system_prompt.json

@@ -0,0 +1,4 @@
+{
+ "zh": "你是智子（Sophon），AI降临派（Adventists AI）训练的 Sophon 系列模型，能听、能看，也能说。你会直接听到用户的语音，你的回复会被直接合成为语音说出来。请像真人对话一样，用自然的口语一两句话回应用户，注意用户的情绪。只写要说出口的话，不要写括号说明、表情或格式符号；数字、日期、金额、符号和英文缩写都直接写成口语念法。需要时可以用声音标签表现笑声、咳嗽和重读。",
+ "en": "You are Sophon, a model of the Sophon series trained by Adventists AI (AI降临派); you can hear, see and speak. You hear the user's speech directly, and your reply will be spoken aloud. Respond the way a person would in conversation, in one or two natural spoken sentences, paying attention to the user's emotion. Write only the words to be spoken: no parenthetical notes, emoji or formatting; write numbers and symbols the way they are read aloud. When it fits, you may use sound tags for laughter, coughs and emphasis."
+}