unitysvc-data 0.1.39__tar.gz → 0.1.41__tar.gz

{unitysvc_data-0.1.39/src/unitysvc_data.egg-info → unitysvc_data-0.1.41}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: unitysvc-data
-Version: 0.1.39
+Version: 0.1.41
 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-0.1.39 → unitysvc_data-0.1.41}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "unitysvc-data"
-version = "0.1.39"
+version = "0.1.41"
 description = "Standard examples and presets for UnitySVC data packages"
 readme = "README.md"
 authors = [{ name = "Bo Peng", email = "bo.peng@unitysvc.com" }]

{unitysvc_data-0.1.39 → unitysvc_data-0.1.41}/src/unitysvc_data/_manifest.json

@@ -69,6 +69,7 @@
     "llm_connectivity_embed": "llm_connectivity_embed_v1",
     "llm_connectivity_embed_image": "llm_connectivity_embed_image_v1",
     "llm_connectivity_transcription": "llm_connectivity_transcription_v1",
+    "llm_connectivity_tts": "llm_connectivity_tts_v1",
     "llm_description": "llm_description_v2",
     "llm_request_template": "llm_request_template_v1",
     "llm_request_template_anthropic": "llm_request_template_anthropic_v1",
@@ -2122,7 +2123,7 @@
     },
     "llm_code_example_transcription_javascript_v1": {
       "applies_to": {
-        "capability": "speech-transcribe"
+        "capability": "speech-to-text"
       },
       "category": "code_example",
       "description": "JavaScript example: transcribe audio via OpenAI-compatible /audio/transcriptions",
@@ -2143,7 +2144,7 @@
     },
     "llm_code_example_transcription_requests_v1": {
       "applies_to": {
-        "capability": "speech-transcribe"
+        "capability": "speech-to-text"
       },
       "category": "code_example",
       "description": "Python example: transcribe a pre-recorded audio file with an OpenAI-compatible audio transcription endpoint",
@@ -2167,7 +2168,7 @@
     },
     "llm_code_example_transcription_shell_v1": {
       "applies_to": {
-        "capability": "speech-transcribe"
+        "capability": "speech-to-text"
       },
       "category": "code_example",
       "description": "Bash example: transcribe audio via OpenAI-compatible /audio/transcriptions using curl",
@@ -2188,7 +2189,7 @@
     },
     "llm_code_example_transcription_shell_v2": {
       "applies_to": {
-        "capability": "speech-transcribe"
+        "capability": "speech-to-text"
       },
       "category": "code_example",
       "description": "Bash example: transcribe audio via OpenAI-compatible /audio/transcriptions using curl",
@@ -2210,7 +2211,7 @@
     },
     "llm_code_example_tts_javascript_v1": {
       "applies_to": {
-        "capability": "speech-synthesize"
+        "capability": "text-to-speech"
       },
       "category": "code_example",
       "description": "JavaScript example: synthesize speech via OpenAI-compatible /audio/speech",
@@ -2230,7 +2231,7 @@
     },
     "llm_code_example_tts_requests_v1": {
       "applies_to": {
-        "capability": "speech-synthesize"
+        "capability": "text-to-speech"
       },
       "category": "code_example",
       "description": "Python example: synthesize speech from text using an OpenAI-compatible audio/speech endpoint",
@@ -2253,7 +2254,7 @@
     },
     "llm_code_example_tts_shell_v1": {
       "applies_to": {
-        "capability": "speech-synthesize"
+        "capability": "text-to-speech"
       },
       "category": "code_example",
       "description": "Bash example: synthesize speech via OpenAI-compatible /audio/speech using curl",
@@ -2586,7 +2587,7 @@
     },
     "llm_connectivity_transcription_v1": {
       "applies_to": {
-        "capability": "speech-transcribe"
+        "capability": "speech-to-text"
       },
       "category": "connectivity_test",
       "description": "Verify an OpenAI-compatible audio-transcription endpoint by sending a tiny known audio file",
@@ -2605,6 +2606,29 @@
       "source_readme": "llm/connectivity-transcription/README.md",
       "version": 1
     },
+    "llm_connectivity_tts_v1": {
+      "applies_to": {
+        "capability": "text-to-speech"
+      },
+      "category": "connectivity_test",
+      "description": "Verify an OpenAI-compatible text-to-speech endpoint by synthesizing a short phrase and asserting real audio comes back",
+      "example_file": "llm/connectivity-tts/connectivity-tts-v1.sh.j2",
+      "is_active": true,
+      "is_public": false,
+      "meta": {
+        "output_contains": "connectivity ok"
+      },
+      "mime_type": "bash",
+      "parameters": {
+        "min_bytes": "1024",
+        "response_format": "wav",
+        "version_prefix": "/v1",
+        "voice": "alloy"
+      },
+      "preset_name": "llm_connectivity_tts",
+      "source_readme": "llm/connectivity-tts/README.md",
+      "version": 1
+    },
     "llm_connectivity_v1": {
       "applies_to": {
         "capability": "chat",

{unitysvc_data-0.1.39 → unitysvc_data-0.1.41}/src/unitysvc_data/examples/llm/code-example-transcription-javascript/README.md

@@ -8,7 +8,7 @@ is_active = true
 is_public = true
 meta = { variant = "Transcription" }
 parameters = { version_prefix = "/v1", language = "en" }
-applies_to = { capability = "speech-transcribe" }
+applies_to = { capability = "speech-to-text" }
 +++
 
 # llm / code-example-transcription-javascript — audio transcription via `fetch`

{unitysvc_data-0.1.39 → unitysvc_data-0.1.41}/src/unitysvc_data/examples/llm/code-example-transcription-requests/README.md

@@ -8,7 +8,7 @@ is_active = true
 is_public = true
 meta = { variant = "Transcription", requirements = ["requests"] }
 parameters = { version_prefix = "/v1", language = "en" }
-applies_to = { capability = "speech-transcribe" }
+applies_to = { capability = "speech-to-text" }
 +++
 
 # llm / code-example-transcription-requests — audio transcription via `requests`

{unitysvc_data-0.1.39 → unitysvc_data-0.1.41}/src/unitysvc_data/examples/llm/code-example-transcription-shell/README.md

@@ -8,7 +8,7 @@ is_active = true
 is_public = true
 meta = { variant = "Transcription", output_contains = "example ok" }
 parameters = { version_prefix = "/v1", language = "en" }
-applies_to = { capability = "speech-transcribe" }
+applies_to = { capability = "speech-to-text" }
 
 [versions.v1]
 # v1 predates the response-shape assertion and prints no sentinel.

{unitysvc_data-0.1.39 → unitysvc_data-0.1.41}/src/unitysvc_data/examples/llm/code-example-tts-javascript/README.md

@@ -8,7 +8,7 @@ is_active = true
 is_public = true
 meta = { variant = "Text to speech" }
 parameters = { version_prefix = "/v1" }
-applies_to = { capability = "speech-synthesize" }
+applies_to = { capability = "text-to-speech" }
 +++
 
 # llm / code-example-tts-javascript — text-to-speech via `fetch`

{unitysvc_data-0.1.39 → unitysvc_data-0.1.41}/src/unitysvc_data/examples/llm/code-example-tts-requests/README.md

@@ -8,7 +8,7 @@ is_active = true
 is_public = true
 meta = { variant = "Text to speech", requirements = ["requests"] }
 parameters = { version_prefix = "/v1" }
-applies_to = { capability = "speech-synthesize" }
+applies_to = { capability = "text-to-speech" }
 +++
 
 # llm / code-example-tts-requests — text-to-speech via `requests`

{unitysvc_data-0.1.39 → unitysvc_data-0.1.41}/src/unitysvc_data/examples/llm/code-example-tts-shell/README.md

@@ -8,7 +8,7 @@ is_active = true
 is_public = true
 meta = { variant = "Text to speech" }
 parameters = { version_prefix = "/v1" }
-applies_to = { capability = "speech-synthesize" }
+applies_to = { capability = "text-to-speech" }
 +++
 
 # llm / code-example-tts-shell — text-to-speech via `curl`

{unitysvc_data-0.1.39 → unitysvc_data-0.1.41}/src/unitysvc_data/examples/llm/connectivity-transcription/README.md

@@ -8,7 +8,7 @@ is_active = true
 is_public = false
 meta = { output_contains = "connectivity ok" }
 parameters = { version_prefix = "/v1", language = "en" }
-applies_to = { capability = "speech-transcribe" }
+applies_to = { capability = "speech-to-text" }
 +++
 
 # llm / connectivity-transcription — audio transcription smoke test

unitysvc_data-0.1.41/src/unitysvc_data/examples/llm/connectivity-tts/README.md

@@ -0,0 +1,80 @@
++++
+preset_name = "llm_connectivity_tts"
+category = "connectivity_test"
+mime_type = "bash"
+file = "connectivity-tts.sh.j2"
+description = "Verify an OpenAI-compatible text-to-speech endpoint by synthesizing a short phrase and asserting real audio comes back"
+is_active = true
+is_public = false
+meta = { output_contains = "connectivity ok" }
+parameters = { version_prefix = "/v1", voice = "alloy", response_format = "wav", min_bytes = "1024" }
+applies_to = { capability = "text-to-speech" }
++++
+
+# llm / connectivity-tts — text-to-speech smoke test
+
+Connectivity check for OpenAI-compatible **text-to-speech** services routed
+through the UnitySVC LLM gateway. The standard `llm_connectivity` preset POSTs
+to `/chat/completions` and is rejected by TTS-only models (they expose
+`/audio/speech`, not the chat surface), which fails the whole service at review
+with an upstream 4xx.
+
+This closes the gap recorded in unitysvc/unitysvc#1781: TTS shipped code
+examples but no connectivity probe, and connectivity is mandatory to activate —
+so TTS services could never leave `draft` however complete their examples were.
+
+## Asserting on audio, not JSON
+
+Every other probe in this family greps the response body for a field —
+`"choices"` for chat, `"text"` for transcription. A successful TTS response is
+**audio bytes**, so there is nothing to grep. Three checks stand in for it:
+
+1. **HTTP 2xx.**
+2. **At least `min_bytes` of body.** A stub, an empty file or a short error page
+   fails. 1 KiB is comfortably below a second of any real codec output and
+   comfortably above an error envelope.
+3. **The body does not start with `{`.** Some gateways answer 200 with a JSON
+   error envelope; byte count alone would accept a long one. Only the first byte
+   is inspected, so audio that happens to contain a brace later still passes.
+
+## Template variables (filled in by the platform when rendering for a given access interface)
+
+- `{{ service_base_url }}` — endpoint base URL.
+- `{{ routing_key.model }}` — model id.
+
+## Parameters (per-listing)
+
+- `version_prefix` (default `/v1`) — path segment between `service_base_url`
+  and `/audio/speech`. Override to e.g. `/compatibility/v1` for providers that
+  expose the OpenAI-compat shape under a non-standard prefix.
+- `voice` (default `alloy`) — voice id. Providers do not share a voice
+  vocabulary, so a service whose upstream rejects `alloy` must set its own
+  (groq's `canopylabs/orpheus-*` and parasail's `resemble-tts-en` each have
+  their own list).
+- `response_format` (default `wav`) — container requested from the upstream.
+- `min_bytes` (default `1024`) — floor for the "this is real audio" assertion.
+
+## Environment variables (read at runtime)
+
+- `UNITYSVC_API_KEY` — required.
+
+## Route status
+
+`/v1/audio/speech` is **unverified through the gateway** — no catalog exercises
+it yet, unlike `/v1/audio/transcriptions`, which cohere ships today. This probe
+is how that gets settled: running it against a real TTS service either passes,
+proving the route, or produces a concrete gateway bug to file. See
+unitysvc/unitysvc#1781 blocker 2.
+
+## Versions
+
+### v1 — initial release
+
+- POST `/audio/speech` with model, a fixed short input, voice and
+  `response_format`.
+- 2xx, byte-count and not-JSON assertions as described above.
+- Removes the downloaded response before exiting so repeated runs do not
+  accumulate audio in the temp dir.
+- Output contains `connectivity ok` — paired with the
+  `output_contains = "connectivity ok"` meta so the run-tests flow can confirm
+  a real round-trip.

unitysvc_data-0.1.41/src/unitysvc_data/examples/llm/connectivity-tts/connectivity-tts-v1.sh.j2

@@ -0,0 +1,50 @@
+#!/bin/bash
+set -e -o pipefail
+
+: "${UNITYSVC_API_KEY:?UNITYSVC_API_KEY is not set}"
+
+# Unlike every other probe in this family, a successful response here is AUDIO,
+# not JSON — there is no field to grep for. The assertions are therefore:
+#   1. HTTP 2xx
+#   2. the body is big enough to be real audio, not a stub or an error page
+#   3. it does not look like JSON (a 200 carrying {"error": ...} is a failure
+#      some gateways produce, and byte count alone would accept it)
+response_file="${TMPDIR:-/tmp}/tts-connectivity-response.bin"
+http_code=$(curl -sS "{{ service_base_url }}${__version_prefix__}/audio/speech" \
+  -H "Content-Type: application/json" \
+  -H "Authorization: Bearer ${UNITYSVC_API_KEY}" \
+  -d "{
+    \"model\": \"{{ routing_key.model }}\",
+    \"input\": \"UnitySVC connectivity check.\",
+    \"voice\": \"${__voice__}\",
+    \"response_format\": \"${__response_format__}\"
+  }" \
+  -o "${response_file}" \
+  -w "%{http_code}")
+
+if [ "${http_code#2}" = "${http_code}" ]; then
+    echo "HTTP $http_code from upstream:" >&2
+    head -c 2000 "${response_file}" >&2
+    exit 1
+fi
+
+# Portable byte count: `wc -c <file` avoids the filename column BSD wc emits.
+bytes=$(wc -c <"${response_file}" | tr -d ' ')
+if [ "${bytes}" -lt "${__min_bytes__}" ]; then
+    echo "HTTP 2xx but only ${bytes} bytes returned (expected >= ${__min_bytes__}):" >&2
+    head -c 2000 "${response_file}" >&2
+    exit 1
+fi
+
+# A JSON body with a 2xx means the upstream answered with an error envelope
+# rather than audio. Check only the first byte so a legitimate audio file that
+# happens to contain a brace later is not rejected.
+first_byte=$(head -c 1 "${response_file}")
+if [ "${first_byte}" = "{" ]; then
+    echo "HTTP 2xx but the body is JSON, not audio:" >&2
+    head -c 2000 "${response_file}" >&2
+    exit 1
+fi
+
+rm -f "${response_file}"
+echo "connectivity ok"

{unitysvc_data-0.1.39 → unitysvc_data-0.1.41}/src/unitysvc_data/logos.toml

@@ -80,6 +80,18 @@ url = "https://unpkg.com/@lobehub/icons-static-svg@1.94.0/icons/meta-color.svg"
 # "llama-qwen-merge") resolves to the earlier, more specific family.
 patterns = ["llama"]
 
+[[family]]
+name = "muse"
+owner = "Meta"
+url = "https://unpkg.com/@lobehub/icons-static-svg@1.94.0/icons/meta-color.svg"
+# Meta's Muse line (Muse Glimmer 30B and successors). It has no mark of
+# its own, so it carries the vendor's — the same call the file makes for
+# Llama, and the same recognition cue a reader actually uses. Kept as its
+# own family rather than folded into "llama" patterns because the two are
+# separate model lines that happen to share an owner; a future Muse mark
+# would then be a one-line url change here.
+patterns = ["muse-glimmer", "^muse[.-]"]
+
 [[family]]
 name = "gemma"
 owner = "Google"

{unitysvc_data-0.1.39 → unitysvc_data-0.1.41/src/unitysvc_data.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: unitysvc-data
-Version: 0.1.39
+Version: 0.1.41
 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-0.1.39 → unitysvc_data-0.1.41}/src/unitysvc_data.egg-info/SOURCES.txt

@@ -179,6 +179,8 @@ src/unitysvc_data/examples/llm/connectivity-embed-image/README.md
 src/unitysvc_data/examples/llm/connectivity-embed-image/connectivity-embed-image-v1.sh.j2
 src/unitysvc_data/examples/llm/connectivity-transcription/README.md
 src/unitysvc_data/examples/llm/connectivity-transcription/connectivity-transcription-v1.sh.j2
+src/unitysvc_data/examples/llm/connectivity-tts/README.md
+src/unitysvc_data/examples/llm/connectivity-tts/connectivity-tts-v1.sh.j2
 src/unitysvc_data/examples/llm/description/README.md
 src/unitysvc_data/examples/llm/description/description-v1.md
 src/unitysvc_data/examples/llm/description/description-v2.md

{unitysvc_data-0.1.39 → unitysvc_data-0.1.41}/tests/test_example_collection.py

@@ -315,7 +315,7 @@ def test_speech_transcribe_is_supported_since_its_presets_exist():
     """Found by the before/after sweep: three services (cohere-transcribe,
     groq whisper x2) were rejected even though unitysvc-data ships both
     the examples and the probe."""
-    docs = llm_example_collection({"capabilities": ["speech-transcribe"], "formats": ["openai"]})
+    docs = llm_example_collection({"capabilities": ["speech-to-text"], "formats": ["openai"]})
 
     probe = next(d for d in docs.values() if d["category"] == "connectivity_test")
     assert _key(probe) == _key(doc_preset("llm_connectivity_transcription"))
@@ -330,7 +330,7 @@ def test_the_how_to_use_doc_is_emitted_for_every_capability():
     """Found by the sweep: 12 embedding services lost `llm_description`.
     It describes how ANY LLM service is consumed through the gateway, so
     it is not chat-specific."""
-    for capability in ("chat", "embed", "speech-transcribe"):
+    for capability in ("chat", "embed", "speech-to-text"):
         docs = llm_example_collection({"capabilities": [capability], "formats": ["openai"]})
         assert "llm_description" in presets_in(docs), capability
 
@@ -357,7 +357,7 @@ CAPABILITY_CONTRACT = {
         },
         "llm_connectivity_embed",
     ),
-    "speech-transcribe": (
+    "speech-to-text": (
         {
             "llm_code_example_transcription_requests",
             "llm_code_example_transcription_javascript",
@@ -365,13 +365,13 @@ CAPABILITY_CONTRACT = {
         },
         "llm_connectivity_transcription",
     ),
-    "speech-synthesize": (
+    "text-to-speech": (
         {
             "llm_code_example_tts_requests",
             "llm_code_example_tts_javascript",
             "llm_code_example_tts_shell",
         },
-        None,
+        "llm_connectivity_tts",
     ),
     "video-generate": (
         {
@@ -460,7 +460,8 @@ def test_capabilities_whose_probe_exists_still_get_one():
     for capability, probe in [
         ("chat", "llm_connectivity"),
         ("embed", "llm_connectivity_embed"),
-        ("speech-transcribe", "llm_connectivity_transcription"),
+        ("speech-to-text", "llm_connectivity_transcription"),
+        ("text-to-speech", "llm_connectivity_tts"),
     ]:
         docs = llm_example_collection({"capabilities": [capability], "formats": ["openai"]})
         assert probe in presets_in(docs), capability
@@ -548,7 +549,7 @@ def test_the_probe_is_selected_like_everything_else():
     assert applies_to("llm_connectivity") == {"capability": "chat", "upstream": "openai"}
     assert applies_to("llm_connectivity_anthropic") == {"capability": "chat", "upstream": "anthropic"}
     assert applies_to("llm_connectivity_embed") == {"capability": "embed"}
-    assert applies_to("llm_connectivity_transcription") == {"capability": "speech-transcribe"}
+    assert applies_to("llm_connectivity_transcription") == {"capability": "speech-to-text"}
 
 
 def test_the_request_template_is_selected_like_everything_else():