@miller-tech/uap 1.220.11 → 1.221.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/config/llama-profiles/gemma4-26b-a4b-mtp.env +9 -0
- package/config/llama-profiles/qwen36-35b-a3b.env +9 -0
- package/config/llama-profiles/qwen38-27b-dflash2.env +9 -0
- package/config/llama-profiles/qwen38-27b-mtp.env +9 -0
- package/dist/.tsbuildinfo +1 -1
- package/dist/delivery/visual-gate.d.ts +16 -1
- package/dist/delivery/visual-gate.d.ts.map +1 -1
- package/dist/delivery/visual-gate.js +14 -3
- package/dist/delivery/visual-gate.js.map +1 -1
- package/package.json +2 -2
- package/src/policies/enforcers/__pycache__/_common.cpython-312.pyc +0 -0
- package/templates/hooks/__pycache__/deliver_autoroute.cpython-312.pyc +0 -0
- package/tools/agents/scripts/__pycache__/toolcall_path_normalizer.cpython-312.pyc +0 -0
- package/tools/agents/scripts/anthropic_proxy.py +116 -8
- package/tools/agents/tests/test_upstream_model_aliases.py +116 -0
|
@@ -41,6 +41,15 @@
|
|
|
41
41
|
# =====================================================================
|
|
42
42
|
LLAMA_BIN=/home/cogtek/llama.cpp/.worktrees/latest-master/build/bin/llama-server
|
|
43
43
|
LLAMA_MODEL=/home/cogtek/Downloads/gemma-4-26B-A4B-it-UD-IQ4_XS.gguf
|
|
44
|
+
# Model id advertised to API clients (llama-server --alias). Set it EXPLICITLY
|
|
45
|
+
# here: the script's fallback derives a name from the GGUF filename, which is a
|
|
46
|
+
# name rather than a path but is not the name client configs are written
|
|
47
|
+
# against ("Qwen3.8-27B-UD-IQ4_XS" vs "Qwen3.8-27B"), so the proxy would still
|
|
48
|
+
# rewrite the model on every request.
|
|
49
|
+
# ORDER IS NOT PRESERVED: llama.cpp stores aliases in a std::set and reports
|
|
50
|
+
# *begin() as the model id, so the ASCII-FIRST entry becomes the primary id.
|
|
51
|
+
# Keep the id clients actually use sorting first if you add more.
|
|
52
|
+
LLAMA_ALIAS=Gemma4-26B-A4B
|
|
44
53
|
LLAMA_DRAFT_MODEL=/home/cogtek/Downloads/mtp-gemma-4-26B-A4B-it-Q8_0.gguf
|
|
45
54
|
|
|
46
55
|
# Vision projector MUST be pinned. The launch script's auto-discovery falls
|
|
@@ -23,6 +23,15 @@
|
|
|
23
23
|
# =====================================================================
|
|
24
24
|
LLAMA_BIN=/home/cogtek/llama.cpp/.worktrees/latest-master/build/bin/llama-server
|
|
25
25
|
LLAMA_MODEL=/home/cogtek/Downloads/Qwen3.6-35B-A3B-UD-IQ4_XS.gguf
|
|
26
|
+
# Model id advertised to API clients (llama-server --alias). Set it EXPLICITLY
|
|
27
|
+
# here: the script's fallback derives a name from the GGUF filename, which is a
|
|
28
|
+
# name rather than a path but is not the name client configs are written
|
|
29
|
+
# against ("Qwen3.8-27B-UD-IQ4_XS" vs "Qwen3.8-27B"), so the proxy would still
|
|
30
|
+
# rewrite the model on every request.
|
|
31
|
+
# ORDER IS NOT PRESERVED: llama.cpp stores aliases in a std::set and reports
|
|
32
|
+
# *begin() as the model id, so the ASCII-FIRST entry becomes the primary id.
|
|
33
|
+
# Keep the id clients actually use sorting first if you add more.
|
|
34
|
+
LLAMA_ALIAS=Qwen3.6-35B-A3B
|
|
26
35
|
LLAMA_MMPROJ=/home/cogtek/Downloads/mmproj-F16.gguf
|
|
27
36
|
|
|
28
37
|
LLAMA_HOST=0.0.0.0
|
|
@@ -68,6 +68,15 @@
|
|
|
68
68
|
# =====================================================================
|
|
69
69
|
LLAMA_BIN=/home/cogtek/llama.cpp/.worktrees/dflash2-trial/build-cuda/bin/llama-server
|
|
70
70
|
LLAMA_MODEL=/home/cogtek/Downloads/Qwen3.8-27B-UD-IQ4_XS.gguf
|
|
71
|
+
# Model id advertised to API clients (llama-server --alias). Set it EXPLICITLY
|
|
72
|
+
# here: the script's fallback derives a name from the GGUF filename, which is a
|
|
73
|
+
# name rather than a path but is not the name client configs are written
|
|
74
|
+
# against ("Qwen3.8-27B-UD-IQ4_XS" vs "Qwen3.8-27B"), so the proxy would still
|
|
75
|
+
# rewrite the model on every request.
|
|
76
|
+
# ORDER IS NOT PRESERVED: llama.cpp stores aliases in a std::set and reports
|
|
77
|
+
# *begin() as the model id, so the ASCII-FIRST entry becomes the primary id.
|
|
78
|
+
# Keep the id clients actually use sorting first if you add more.
|
|
79
|
+
LLAMA_ALIAS=Qwen3.8-27B
|
|
71
80
|
# CONFIRMED INCOMPATIBLE 2026-08-20: mmproj-BF16.gguf crashed the server on
|
|
72
81
|
# load -- "mismatch between text model (n_embd = 5120) and mmproj
|
|
73
82
|
# (n_embd = 2048)". It was built for a different, smaller model's vision
|
|
@@ -36,6 +36,15 @@
|
|
|
36
36
|
# =====================================================================
|
|
37
37
|
LLAMA_BIN=/home/cogtek/llama.cpp/.worktrees/dflash-mtp-upgrade/build-cuda/bin/llama-server
|
|
38
38
|
LLAMA_MODEL=/home/cogtek/Downloads/Qwen3.8-27B-UD-IQ4_XS.gguf
|
|
39
|
+
# Model id advertised to API clients (llama-server --alias). Set it EXPLICITLY
|
|
40
|
+
# here: the script's fallback derives a name from the GGUF filename, which is a
|
|
41
|
+
# name rather than a path but is not the name client configs are written
|
|
42
|
+
# against ("Qwen3.8-27B-UD-IQ4_XS" vs "Qwen3.8-27B"), so the proxy would still
|
|
43
|
+
# rewrite the model on every request.
|
|
44
|
+
# ORDER IS NOT PRESERVED: llama.cpp stores aliases in a std::set and reports
|
|
45
|
+
# *begin() as the model id, so the ASCII-FIRST entry becomes the primary id.
|
|
46
|
+
# Keep the id clients actually use sorting first if you add more.
|
|
47
|
+
LLAMA_ALIAS=Qwen3.8-27B
|
|
39
48
|
# CONFIRMED INCOMPATIBLE 2026-08-20: mmproj-BF16.gguf crashed the server on
|
|
40
49
|
# load -- "mismatch between text model (n_embd = 5120) and mmproj
|
|
41
50
|
# (n_embd = 2048)". It was built for a different, smaller model's vision
|