unitysvc-data 0.2.5__tar.gz → 0.2.6__tar.gz

{unitysvc_data-0.2.5/src/unitysvc_data.egg-info → unitysvc_data-0.2.6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: unitysvc-data
-Version: 0.2.5
+Version: 0.2.6
 Summary: Standard examples and presets for UnitySVC data packages
 Author-email: Bo Peng <bo.peng@unitysvc.com>
 Maintainer-email: Bo Peng <bo.peng@unitysvc.com>

{unitysvc_data-0.2.5 → unitysvc_data-0.2.6}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "unitysvc-data"
-version = "0.2.5"
+version = "0.2.6"
 description = "Standard examples and presets for UnitySVC data packages"
 readme = "README.md"
 authors = [{ name = "Bo Peng", email = "bo.peng@unitysvc.com" }]

{unitysvc_data-0.2.5 → unitysvc_data-0.2.6}/src/unitysvc_data/_manifest.json

@@ -46,6 +46,12 @@
     "llm_code_example_imagetoimage_requests": "llm_code_example_imagetoimage_requests_v1",
     "llm_code_example_imagetoimage_shell": "llm_code_example_imagetoimage_shell_v2",
     "llm_code_example_javascript": "llm_code_example_javascript_v1",
+    "llm_code_example_omni_asr_dashscope_javascript": "llm_code_example_omni_asr_dashscope_javascript_v1",
+    "llm_code_example_omni_asr_dashscope_requests": "llm_code_example_omni_asr_dashscope_requests_v1",
+    "llm_code_example_omni_asr_dashscope_shell": "llm_code_example_omni_asr_dashscope_shell_v1",
+    "llm_code_example_omni_tts_dashscope_javascript": "llm_code_example_omni_tts_dashscope_javascript_v1",
+    "llm_code_example_omni_tts_dashscope_requests": "llm_code_example_omni_tts_dashscope_requests_v1",
+    "llm_code_example_omni_tts_dashscope_shell": "llm_code_example_omni_tts_dashscope_shell_v1",
     "llm_code_example_openai": "llm_code_example_openai_v1",
     "llm_code_example_openai_javascript": "llm_code_example_openai_javascript_v1",
     "llm_code_example_openai_to_anthropic_requests": "llm_code_example_openai_to_anthropic_requests_v1",
@@ -87,6 +93,8 @@
     "llm_connectivity_embed": "llm_connectivity_embed_v1",
     "llm_connectivity_embed_dashscope": "llm_connectivity_embed_dashscope_v1",
     "llm_connectivity_embed_image": "llm_connectivity_embed_image_v1",
+    "llm_connectivity_omni_asr_dashscope": "llm_connectivity_omni_asr_dashscope_v1",
+    "llm_connectivity_omni_tts_dashscope": "llm_connectivity_omni_tts_dashscope_v1",
     "llm_connectivity_transcription": "llm_connectivity_transcription_v1",
     "llm_connectivity_tts": "llm_connectivity_tts_v1",
     "llm_connectivity_tts_dashscope": "llm_connectivity_tts_dashscope_v1",
@@ -1162,7 +1170,7 @@
     "llm_code_example_asr_dashscope_javascript_v1": {
       "applies_to": {
         "capability": "speech-to-text",
-        "dialect": "dashscope_multimodal",
+        "dialect": "dashscope_audio_task",
         "upstream": "dashscope"
       },
       "category": "code_example",
@@ -1186,7 +1194,7 @@
     "llm_code_example_asr_dashscope_requests_v1": {
       "applies_to": {
         "capability": "speech-to-text",
-        "dialect": "dashscope_multimodal",
+        "dialect": "dashscope_audio_task",
         "upstream": "dashscope"
       },
       "category": "code_example",
@@ -1213,7 +1221,7 @@
     "llm_code_example_asr_dashscope_shell_v1": {
       "applies_to": {
         "capability": "speech-to-text",
-        "dialect": "dashscope_multimodal",
+        "dialect": "dashscope_audio_task",
         "upstream": "dashscope"
       },
       "category": "code_example",
@@ -2108,6 +2116,156 @@
       "source_readme": "llm/code-example-javascript/README.md",
       "version": 1
     },
+    "llm_code_example_omni_asr_dashscope_javascript_v1": {
+      "applies_to": {
+        "capability": "speech-to-text",
+        "dialect": "dashscope_multimodal",
+        "upstream": "dashscope"
+      },
+      "category": "code_example",
+      "description": "JavaScript example: transcribe audio with an omni model on a DashScope-native endpoint with fetch",
+      "example_file": "llm/dashscope-omni-asr-javascript/code-example-v1.js.j2",
+      "is_active": true,
+      "is_public": true,
+      "meta": {
+        "min_expected_metrics": {
+          "input_tokens": 1,
+          "output_tokens": 1
+        },
+        "variant": "Transcription (DashScope omni)"
+      },
+      "mime_type": "javascript",
+      "parameters": {},
+      "preset_name": "llm_code_example_omni_asr_dashscope_javascript",
+      "source_readme": "llm/dashscope-omni-asr-javascript/README.md",
+      "version": 1
+    },
+    "llm_code_example_omni_asr_dashscope_requests_v1": {
+      "applies_to": {
+        "capability": "speech-to-text",
+        "dialect": "dashscope_multimodal",
+        "upstream": "dashscope"
+      },
+      "category": "code_example",
+      "description": "Python example: transcribe audio with an omni model on a DashScope-native endpoint using the requests library",
+      "example_file": "llm/dashscope-omni-asr-requests/code-example-v1.py.j2",
+      "is_active": true,
+      "is_public": true,
+      "meta": {
+        "min_expected_metrics": {
+          "input_tokens": 1,
+          "output_tokens": 1
+        },
+        "requirements": [
+          "requests"
+        ],
+        "variant": "Transcription (DashScope omni)"
+      },
+      "mime_type": "python",
+      "parameters": {},
+      "preset_name": "llm_code_example_omni_asr_dashscope_requests",
+      "source_readme": "llm/dashscope-omni-asr-requests/README.md",
+      "version": 1
+    },
+    "llm_code_example_omni_asr_dashscope_shell_v1": {
+      "applies_to": {
+        "capability": "speech-to-text",
+        "dialect": "dashscope_multimodal",
+        "upstream": "dashscope"
+      },
+      "category": "code_example",
+      "description": "Shell example: transcribe audio with an omni model on a DashScope-native endpoint via curl",
+      "example_file": "llm/dashscope-omni-asr-shell/code-example-v1.sh.j2",
+      "is_active": true,
+      "is_public": true,
+      "meta": {
+        "min_expected_metrics": {
+          "input_tokens": 1,
+          "output_tokens": 1
+        },
+        "variant": "Transcription (DashScope omni)"
+      },
+      "mime_type": "bash",
+      "parameters": {},
+      "preset_name": "llm_code_example_omni_asr_dashscope_shell",
+      "source_readme": "llm/dashscope-omni-asr-shell/README.md",
+      "version": 1
+    },
+    "llm_code_example_omni_tts_dashscope_javascript_v1": {
+      "applies_to": {
+        "capability": "text-to-speech",
+        "dialect": "dashscope_multimodal",
+        "upstream": "dashscope"
+      },
+      "category": "code_example",
+      "description": "JavaScript example: synthesise speech with an omni model on a DashScope-native endpoint with fetch",
+      "example_file": "llm/dashscope-omni-tts-javascript/code-example-v1.js.j2",
+      "is_active": true,
+      "is_public": true,
+      "meta": {
+        "min_expected_metrics": {
+          "input_tokens": 1,
+          "output_tokens": 1
+        },
+        "variant": "Speech (DashScope omni)"
+      },
+      "mime_type": "javascript",
+      "parameters": {},
+      "preset_name": "llm_code_example_omni_tts_dashscope_javascript",
+      "source_readme": "llm/dashscope-omni-tts-javascript/README.md",
+      "version": 1
+    },
+    "llm_code_example_omni_tts_dashscope_requests_v1": {
+      "applies_to": {
+        "capability": "text-to-speech",
+        "dialect": "dashscope_multimodal",
+        "upstream": "dashscope"
+      },
+      "category": "code_example",
+      "description": "Python example: synthesise speech with an omni model on a DashScope-native endpoint using the requests library",
+      "example_file": "llm/dashscope-omni-tts-requests/code-example-v1.py.j2",
+      "is_active": true,
+      "is_public": true,
+      "meta": {
+        "min_expected_metrics": {
+          "input_tokens": 1,
+          "output_tokens": 1
+        },
+        "requirements": [
+          "requests"
+        ],
+        "variant": "Speech (DashScope omni)"
+      },
+      "mime_type": "python",
+      "parameters": {},
+      "preset_name": "llm_code_example_omni_tts_dashscope_requests",
+      "source_readme": "llm/dashscope-omni-tts-requests/README.md",
+      "version": 1
+    },
+    "llm_code_example_omni_tts_dashscope_shell_v1": {
+      "applies_to": {
+        "capability": "text-to-speech",
+        "dialect": "dashscope_multimodal",
+        "upstream": "dashscope"
+      },
+      "category": "code_example",
+      "description": "Shell example: synthesise speech with an omni model on a DashScope-native endpoint via curl",
+      "example_file": "llm/dashscope-omni-tts-shell/code-example-v1.sh.j2",
+      "is_active": true,
+      "is_public": true,
+      "meta": {
+        "min_expected_metrics": {
+          "input_tokens": 1,
+          "output_tokens": 1
+        },
+        "variant": "Speech (DashScope omni)"
+      },
+      "mime_type": "bash",
+      "parameters": {},
+      "preset_name": "llm_code_example_omni_tts_dashscope_shell",
+      "source_readme": "llm/dashscope-omni-tts-shell/README.md",
+      "version": 1
+    },
     "llm_code_example_openai_javascript_v1": {
       "applies_to": {
         "capability": "chat",
@@ -2865,7 +3023,7 @@
     "llm_code_example_tts_dashscope_javascript_v1": {
       "applies_to": {
         "capability": "text-to-speech",
-        "dialect": "dashscope_multimodal",
+        "dialect": "dashscope_audio_task",
         "upstream": "dashscope"
       },
       "category": "code_example",
@@ -2888,7 +3046,7 @@
     "llm_code_example_tts_dashscope_requests_v1": {
       "applies_to": {
         "capability": "text-to-speech",
-        "dialect": "dashscope_multimodal",
+        "dialect": "dashscope_audio_task",
         "upstream": "dashscope"
       },
       "category": "code_example",
@@ -2914,7 +3072,7 @@
     "llm_code_example_tts_dashscope_shell_v1": {
       "applies_to": {
         "capability": "text-to-speech",
-        "dialect": "dashscope_multimodal",
+        "dialect": "dashscope_audio_task",
         "upstream": "dashscope"
       },
       "category": "code_example",
@@ -3307,7 +3465,7 @@
     "llm_connectivity_asr_dashscope_v1": {
       "applies_to": {
         "capability": "speech-to-text",
-        "dialect": "dashscope_multimodal",
+        "dialect": "dashscope_audio_task",
         "upstream": "dashscope"
       },
       "category": "connectivity_test",
@@ -3441,6 +3599,52 @@
       "source_readme": "llm/connectivity-embed/README.md",
       "version": 1
     },
+    "llm_connectivity_omni_asr_dashscope_v1": {
+      "applies_to": {
+        "capability": "speech-to-text",
+        "dialect": "dashscope_multimodal",
+        "upstream": "dashscope"
+      },
+      "category": "connectivity_test",
+      "description": "Verify an omni DashScope endpoint by transcribing a tiny audio sample",
+      "example_file": "llm/dashscope-omni-asr-connectivity/connectivity-v1.sh.j2",
+      "is_active": true,
+      "is_public": false,
+      "meta": {
+        "min_expected_metrics": {
+          "input_tokens": 1,
+          "output_tokens": 1
+        }
+      },
+      "mime_type": "bash",
+      "parameters": {},
+      "preset_name": "llm_connectivity_omni_asr_dashscope",
+      "source_readme": "llm/dashscope-omni-asr-connectivity/README.md",
+      "version": 1
+    },
+    "llm_connectivity_omni_tts_dashscope_v1": {
+      "applies_to": {
+        "capability": "text-to-speech",
+        "dialect": "dashscope_multimodal",
+        "upstream": "dashscope"
+      },
+      "category": "connectivity_test",
+      "description": "Verify an omni DashScope endpoint by synthesising a short clip",
+      "example_file": "llm/dashscope-omni-tts-connectivity/connectivity-v1.sh.j2",
+      "is_active": true,
+      "is_public": false,
+      "meta": {
+        "min_expected_metrics": {
+          "input_tokens": 1,
+          "output_tokens": 1
+        }
+      },
+      "mime_type": "bash",
+      "parameters": {},
+      "preset_name": "llm_connectivity_omni_tts_dashscope",
+      "source_readme": "llm/dashscope-omni-tts-connectivity/README.md",
+      "version": 1
+    },
     "llm_connectivity_transcription_v1": {
       "applies_to": {
         "capability": "speech-to-text",
@@ -3468,7 +3672,7 @@
     "llm_connectivity_tts_dashscope_v1": {
       "applies_to": {
         "capability": "text-to-speech",
-        "dialect": "dashscope_multimodal",
+        "dialect": "dashscope_audio_task",
         "upstream": "dashscope"
       },
       "category": "connectivity_test",

{unitysvc_data-0.2.5 → unitysvc_data-0.2.6}/src/unitysvc_data/examples/llm/dashscope-asr-connectivity/README.md

@@ -7,7 +7,7 @@ description = "Verify a DashScope-native endpoint by issuing a tiny request (spe
 is_active = true
 is_public = false
 meta = { min_expected_metrics = { input_tokens = 1, output_tokens = 1 } }
-applies_to = { capability = "speech-to-text", dialect = "dashscope_multimodal", upstream = "dashscope" }
+applies_to = { capability = "speech-to-text", dialect = "dashscope_audio_task", upstream = "dashscope" }
 +++
 
 # llm / dashscope-asr-connectivity
@@ -43,3 +43,10 @@ the full native path, so the example posts to the service URL with no suffix.
   `dashscope_text` instead; a bare `dashscope` now selects neither, so a service
   has to say which it speaks. Selection metadata only — nothing these examples do
   has changed.
+- **Amended in 0.2.6**: the dialect token is now `dashscope_audio_task`. This
+  example is written for a DEDICATED single-purpose audio model, whose request is
+  `input.text` / `input.audio` with no `messages`. An omni model rejects that with
+  `InvalidParameter: Either "prompt" or "messages" must exist`, and reaches the
+  same capability as a chat call asking for audio — which is what the new
+  `dashscope-omni-*` families do under `dashscope_multimodal`. Both had claimed
+  that token, so the same capability selected both and one overwrote the other.

{unitysvc_data-0.2.5 → unitysvc_data-0.2.6}/src/unitysvc_data/examples/llm/dashscope-asr-javascript/README.md

@@ -7,7 +7,7 @@ description = "JavaScript example: transcribe audio on a DashScope-native endpoi
 is_active = true
 is_public = true
 meta = { variant = "Speech to text (DashScope)", min_expected_metrics = { input_tokens = 1, output_tokens = 1 } }
-applies_to = { capability = "speech-to-text", dialect = "dashscope_multimodal", upstream = "dashscope" }
+applies_to = { capability = "speech-to-text", dialect = "dashscope_audio_task", upstream = "dashscope" }
 +++
 
 # llm / dashscope-asr-javascript
@@ -43,3 +43,10 @@ the full native path, so the example posts to the service URL with no suffix.
   `dashscope_text` instead; a bare `dashscope` now selects neither, so a service
   has to say which it speaks. Selection metadata only — nothing these examples do
   has changed.
+- **Amended in 0.2.6**: the dialect token is now `dashscope_audio_task`. This
+  example is written for a DEDICATED single-purpose audio model, whose request is
+  `input.text` / `input.audio` with no `messages`. An omni model rejects that with
+  `InvalidParameter: Either "prompt" or "messages" must exist`, and reaches the
+  same capability as a chat call asking for audio — which is what the new
+  `dashscope-omni-*` families do under `dashscope_multimodal`. Both had claimed
+  that token, so the same capability selected both and one overwrote the other.

{unitysvc_data-0.2.5 → unitysvc_data-0.2.6}/src/unitysvc_data/examples/llm/dashscope-asr-requests/README.md

@@ -7,7 +7,7 @@ description = "Python example: transcribe audio on a DashScope-native endpoint u
 is_active = true
 is_public = true
 meta = { variant = "Speech to text (DashScope)", requirements = ["requests"], min_expected_metrics = { input_tokens = 1, output_tokens = 1 } }
-applies_to = { capability = "speech-to-text", dialect = "dashscope_multimodal", upstream = "dashscope" }
+applies_to = { capability = "speech-to-text", dialect = "dashscope_audio_task", upstream = "dashscope" }
 +++
 
 # llm / dashscope-asr-requests
@@ -43,3 +43,10 @@ the full native path, so the example posts to the service URL with no suffix.
   `dashscope_text` instead; a bare `dashscope` now selects neither, so a service
   has to say which it speaks. Selection metadata only — nothing these examples do
   has changed.
+- **Amended in 0.2.6**: the dialect token is now `dashscope_audio_task`. This
+  example is written for a DEDICATED single-purpose audio model, whose request is
+  `input.text` / `input.audio` with no `messages`. An omni model rejects that with
+  `InvalidParameter: Either "prompt" or "messages" must exist`, and reaches the
+  same capability as a chat call asking for audio — which is what the new
+  `dashscope-omni-*` families do under `dashscope_multimodal`. Both had claimed
+  that token, so the same capability selected both and one overwrote the other.

{unitysvc_data-0.2.5 → unitysvc_data-0.2.6}/src/unitysvc_data/examples/llm/dashscope-asr-shell/README.md

@@ -7,7 +7,7 @@ description = "Shell example: transcribe audio on a DashScope-native endpoint vi
 is_active = true
 is_public = true
 meta = { variant = "Speech to text (DashScope)", min_expected_metrics = { input_tokens = 1, output_tokens = 1 } }
-applies_to = { capability = "speech-to-text", dialect = "dashscope_multimodal", upstream = "dashscope" }
+applies_to = { capability = "speech-to-text", dialect = "dashscope_audio_task", upstream = "dashscope" }
 +++
 
 # llm / dashscope-asr-shell
@@ -43,3 +43,10 @@ the full native path, so the example posts to the service URL with no suffix.
   `dashscope_text` instead; a bare `dashscope` now selects neither, so a service
   has to say which it speaks. Selection metadata only — nothing these examples do
   has changed.
+- **Amended in 0.2.6**: the dialect token is now `dashscope_audio_task`. This
+  example is written for a DEDICATED single-purpose audio model, whose request is
+  `input.text` / `input.audio` with no `messages`. An omni model rejects that with
+  `InvalidParameter: Either "prompt" or "messages" must exist`, and reaches the
+  same capability as a chat call asking for audio — which is what the new
+  `dashscope-omni-*` families do under `dashscope_multimodal`. Both had claimed
+  that token, so the same capability selected both and one overwrote the other.

unitysvc_data-0.2.6/src/unitysvc_data/examples/llm/dashscope-omni-asr-connectivity/README.md

@@ -0,0 +1,34 @@
++++
+preset_name = "llm_connectivity_omni_asr_dashscope"
+category = "connectivity_test"
+mime_type = "bash"
+file = "connectivity.sh.j2"
+description = "Verify an omni DashScope endpoint by transcribing a tiny audio sample"
+is_active = true
+is_public = false
+meta = { min_expected_metrics = { input_tokens = 1, output_tokens = 1 } }
+applies_to = { capability = "speech-to-text", dialect = "dashscope_multimodal", upstream = "dashscope" }
++++
+
+# llm / dashscope-omni-asr-connectivity
+
+An **omni** model reaches speech-to-text as a chat call: an `audio` part in
+`input.messages[].content`, alongside the instruction. It is not the
+`input.audio` request a dedicated ASR model takes — an omni model rejects that
+with `InvalidParameter: Either "prompt" or "messages" must exist and cannot both
+be none`. Verified against the live API, which transcribed the sample correctly.
+
+The reply is ordinary chat: the transcription is
+`output.choices[0].message.content[0].text`. Usage bills in **tokens**, with the
+audio counted in `input_tokens_details.audio_tokens` — not in characters.
+
+Assertions check the payload rather than the status, because DashScope answers
+**200 with an error envelope**. They sit behind `{%- if not customer_display %}`
+so the published example stays clean.
+
+## Versions
+
+### v1 — initial release
+
+- Audio by URL in a content part; the model fetches it.
+- Reads the transcription from the chat reply.

unitysvc_data-0.2.6/src/unitysvc_data/examples/llm/dashscope-omni-asr-connectivity/connectivity-v1.sh.j2

@@ -0,0 +1,20 @@
+#!/bin/bash
+# DashScope connectivity probe (omni speech-to-text): transcribe a tiny sample
+# and assert the reply really carries "choices". A 2xx alone is not enough --
+# DashScope answers 200 with an error envelope.
+set -e -o pipefail
+
+: "${UNITYSVC_API_KEY:?UNITYSVC_API_KEY is not set}"
+
+response=$(curl --fail-with-body -sS "{{ service_base_url }}" \
+  -H "Content-Type: application/json" \
+  -H "Authorization: Bearer ${UNITYSVC_API_KEY}" \
+  -d '{"model": "{{ routing_key.model }}", "input": {"messages": [{"role": "user", "content": [{"audio": "https://raw.githubusercontent.com/openai/whisper/main/tests/jfk.flac"}, {"text": "Transcribe this audio."}]}]}, "parameters": {"max_tokens": 64}}')
+
+echo "$response"
+{%- if not customer_display %}
+if ! printf '%s' "$response" | grep -q '"choices"'; then
+  echo 'unexpected response: no "choices" in the reply' >&2
+  exit 1
+fi
+{%- endif %}

unitysvc_data-0.2.6/src/unitysvc_data/examples/llm/dashscope-omni-asr-javascript/README.md

@@ -0,0 +1,34 @@
++++
+preset_name = "llm_code_example_omni_asr_dashscope_javascript"
+category = "code_example"
+mime_type = "javascript"
+file = "code-example.js.j2"
+description = "JavaScript example: transcribe audio with an omni model on a DashScope-native endpoint with fetch"
+is_active = true
+is_public = true
+meta = { variant = "Transcription (DashScope omni)", min_expected_metrics = { input_tokens = 1, output_tokens = 1 } }
+applies_to = { capability = "speech-to-text", dialect = "dashscope_multimodal", upstream = "dashscope" }
++++
+
+# llm / dashscope-omni-asr-javascript
+
+An **omni** model reaches speech-to-text as a chat call: an `audio` part in
+`input.messages[].content`, alongside the instruction. It is not the
+`input.audio` request a dedicated ASR model takes — an omni model rejects that
+with `InvalidParameter: Either "prompt" or "messages" must exist and cannot both
+be none`. Verified against the live API, which transcribed the sample correctly.
+
+The reply is ordinary chat: the transcription is
+`output.choices[0].message.content[0].text`. Usage bills in **tokens**, with the
+audio counted in `input_tokens_details.audio_tokens` — not in characters.
+
+Assertions check the payload rather than the status, because DashScope answers
+**200 with an error envelope**. They sit behind `{%- if not customer_display %}`
+so the published example stays clean.
+
+## Versions
+
+### v1 — initial release
+
+- Audio by URL in a content part; the model fetches it.
+- Reads the transcription from the chat reply.

unitysvc_data-0.2.6/src/unitysvc_data/examples/llm/dashscope-omni-asr-javascript/code-example-v1.js.j2

@@ -0,0 +1,39 @@
+async function main() {
+  const response = await fetch("{{ service_base_url }}", {
+    method: "POST",
+    headers: {
+      "Content-Type": "application/json",
+      Authorization: `Bearer ${process.env.UNITYSVC_API_KEY}`,
+    },
+    body: JSON.stringify({
+      model: "{{ routing_key.model }}",
+      input: {
+        messages: [
+          {
+            role: "user",
+            content: [
+              { audio: "https://raw.githubusercontent.com/openai/whisper/main/tests/jfk.flac" },
+              { text: "Transcribe this audio." },
+            ],
+          },
+        ],
+      },
+      parameters: { max_tokens: 128 },
+    }),
+  });
+  if (!response.ok) {
+    throw new Error(`HTTP ${response.status}: ${await response.text()}`);
+  }
+  const data = await response.json();
+  console.log(data.output.choices[0].message.content[0].text);
+{%- if not customer_display %}
+  if (!data.output?.choices?.length) {
+      throw new Error("no choices in the reply");
+    }
+{%- endif %}
+}
+
+main().catch((err) => {
+  console.error(err.message);
+  process.exit(1);
+});

unitysvc_data-0.2.6/src/unitysvc_data/examples/llm/dashscope-omni-asr-requests/README.md

@@ -0,0 +1,34 @@
++++
+preset_name = "llm_code_example_omni_asr_dashscope_requests"
+category = "code_example"
+mime_type = "python"
+file = "code-example.py.j2"
+description = "Python example: transcribe audio with an omni model on a DashScope-native endpoint using the requests library"
+is_active = true
+is_public = true
+meta = { variant = "Transcription (DashScope omni)", requirements = ["requests"], min_expected_metrics = { input_tokens = 1, output_tokens = 1 } }
+applies_to = { capability = "speech-to-text", dialect = "dashscope_multimodal", upstream = "dashscope" }
++++
+
+# llm / dashscope-omni-asr-requests
+
+An **omni** model reaches speech-to-text as a chat call: an `audio` part in
+`input.messages[].content`, alongside the instruction. It is not the
+`input.audio` request a dedicated ASR model takes — an omni model rejects that
+with `InvalidParameter: Either "prompt" or "messages" must exist and cannot both
+be none`. Verified against the live API, which transcribed the sample correctly.
+
+The reply is ordinary chat: the transcription is
+`output.choices[0].message.content[0].text`. Usage bills in **tokens**, with the
+audio counted in `input_tokens_details.audio_tokens` — not in characters.
+
+Assertions check the payload rather than the status, because DashScope answers
+**200 with an error envelope**. They sit behind `{%- if not customer_display %}`
+so the published example stays clean.
+
+## Versions
+
+### v1 — initial release
+
+- Audio by URL in a content part; the model fetches it.
+- Reads the transcription from the chat reply.

unitysvc_data-0.2.6/src/unitysvc_data/examples/llm/dashscope-omni-asr-requests/code-example-v1.py.j2

@@ -0,0 +1,34 @@
+import os
+
+import requests
+
+response = requests.post(
+    "{{ service_base_url }}",
+    headers={
+        "Authorization": f"Bearer {os.environ['UNITYSVC_API_KEY']}",
+        "Content-Type": "application/json",
+    },
+    json={
+        "model": "{{ routing_key.model }}",
+        "input": {
+            "messages": [
+                {
+                    "role": "user",
+                    "content": [
+                        {"audio": "https://raw.githubusercontent.com/openai/whisper/main/tests/jfk.flac"},
+                        {"text": "Transcribe this audio."},
+                    ],
+                }
+            ]
+        },
+        "parameters": {"max_tokens": 128},
+    },
+)
+if response.status_code >= 400:
+    raise RuntimeError(f"HTTP {response.status_code} from {response.url}: {response.text}")
+data = response.json()
+print(data["output"]["choices"][0]["message"]["content"][0]["text"])
+{%- if not customer_display %}
+if not data.get("output", {}).get("choices"):
+    raise SystemExit("no choices in the reply")
+{%- endif %}

unitysvc_data-0.2.6/src/unitysvc_data/examples/llm/dashscope-omni-asr-shell/README.md

@@ -0,0 +1,34 @@
++++
+preset_name = "llm_code_example_omni_asr_dashscope_shell"
+category = "code_example"
+mime_type = "bash"
+file = "code-example.sh.j2"
+description = "Shell example: transcribe audio with an omni model on a DashScope-native endpoint via curl"
+is_active = true
+is_public = true
+meta = { variant = "Transcription (DashScope omni)", min_expected_metrics = { input_tokens = 1, output_tokens = 1 } }
+applies_to = { capability = "speech-to-text", dialect = "dashscope_multimodal", upstream = "dashscope" }
++++
+
+# llm / dashscope-omni-asr-shell
+
+An **omni** model reaches speech-to-text as a chat call: an `audio` part in
+`input.messages[].content`, alongside the instruction. It is not the
+`input.audio` request a dedicated ASR model takes — an omni model rejects that
+with `InvalidParameter: Either "prompt" or "messages" must exist and cannot both
+be none`. Verified against the live API, which transcribed the sample correctly.
+
+The reply is ordinary chat: the transcription is
+`output.choices[0].message.content[0].text`. Usage bills in **tokens**, with the
+audio counted in `input_tokens_details.audio_tokens` — not in characters.
+
+Assertions check the payload rather than the status, because DashScope answers
+**200 with an error envelope**. They sit behind `{%- if not customer_display %}`
+so the published example stays clean.
+
+## Versions
+
+### v1 — initial release
+
+- Audio by URL in a content part; the model fetches it.
+- Reads the transcription from the chat reply.

unitysvc_data-0.2.6/src/unitysvc_data/examples/llm/dashscope-omni-asr-shell/code-example-v1.sh.j2

@@ -0,0 +1,18 @@
+#!/bin/bash
+set -e -o pipefail
+
+: "${UNITYSVC_API_KEY:?UNITYSVC_API_KEY is not set}"
+
+response=$(curl --fail-with-body -sS "{{ service_base_url }}" \
+  -H "Content-Type: application/json" \
+  -H "Authorization: Bearer ${UNITYSVC_API_KEY}" \
+  -d '{"model": "{{ routing_key.model }}", "input": {"messages": [{"role": "user", "content": [{"audio": "https://raw.githubusercontent.com/openai/whisper/main/tests/jfk.flac"}, {"text": "Transcribe this audio."}]}]}, "parameters": {"max_tokens": 128}}')
+
+echo "$response"
+{%- if not customer_display %}
+# DashScope answers 200 with an error envelope, so check the payload.
+if ! printf '%s' "$response" | grep -q '"choices"'; then
+  echo 'unexpected response: no "choices" in the reply' >&2
+  exit 1
+fi
+{%- endif %}