Merge pull request #405 from janhq/update-dev-from-master-2026-01-27-00-41

jan-service-account · web-flow · commit 29ffd3bf0ebc · 2026-01-27T07:42:26.000+07:00
Sync master with upstream release b7839
diff --git a/.github/workflows/check-vendor.yml b/.github/workflows/check-vendor.yml
@@ -19,7 +19,7 @@ on:
 
 jobs:
   check-vendor:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-slim
 
     steps:
       - name: Checkout
diff --git a/.github/workflows/close-issue.yml b/.github/workflows/close-issue.yml
@@ -10,7 +10,7 @@ permissions:
 
 jobs:
   close-issues:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-slim
     permissions:
       issues: write
       pull-requests: write
diff --git a/.github/workflows/editorconfig.yml b/.github/workflows/editorconfig.yml
@@ -20,7 +20,7 @@ concurrency:
 
 jobs:
   editorconfig:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-slim
     steps:
       - uses: actions/checkout@v6
       - uses: editorconfig-checker/action-editorconfig-checker@v2
diff --git a/.github/workflows/gguf-publish.yml b/.github/workflows/gguf-publish.yml
@@ -21,7 +21,7 @@ on:
 jobs:
   deploy:
 
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-slim
 
     steps:
     - uses: actions/checkout@v6
diff --git a/.github/workflows/labeler.yml b/.github/workflows/labeler.yml
@@ -7,7 +7,7 @@ jobs:
     permissions:
       contents: read
       pull-requests: write
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-slim
     steps:
     - uses: actions/checkout@v6
       with:
diff --git a/.github/workflows/pre-tokenizer-hashes.yml b/.github/workflows/pre-tokenizer-hashes.yml
@@ -12,7 +12,7 @@ on:
 
 jobs:
     pre-tokenizer-hashes:
-        runs-on: ubuntu-latest
+        runs-on: ubuntu-slim
 
         steps:
         - name: Checkout repository
diff --git a/.github/workflows/python-check-requirements.yml b/.github/workflows/python-check-requirements.yml
@@ -20,7 +20,7 @@ concurrency:
 
 jobs:
   python-check-requirements:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-slim
     name: check-requirements
     steps:
       - name: Check out source repository
diff --git a/.github/workflows/python-lint.yml b/.github/workflows/python-lint.yml
@@ -15,7 +15,7 @@ concurrency:
 
 jobs:
   flake8-lint:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-slim
     name: Lint
     steps:
       - name: Check out source repository
diff --git a/.github/workflows/python-type-check.yml b/.github/workflows/python-type-check.yml
@@ -29,9 +29,7 @@ jobs:
         uses: actions/setup-python@v6
         with:
           python-version: "3.11"
-      - name: Install Python dependencies
-        # TODO: use a venv
-        run: pip install -r requirements/requirements-all.txt
+          pip-install: -r requirements/requirements-all.txt
       - name: Type-check with Pyright
         uses: jakebailey/pyright-action@v2
         with:
diff --git a/.github/workflows/update-ops-docs.yml b/.github/workflows/update-ops-docs.yml
@@ -14,7 +14,7 @@ on:
 
 jobs:
     update-ops-docs:
-        runs-on: ubuntu-latest
+        runs-on: ubuntu-slim
 
         steps:
         - name: Checkout repository
diff --git a/.github/workflows/winget.yml b/.github/workflows/winget.yml
@@ -8,7 +8,7 @@ on:
 jobs:
   update:
     name: Update Winget Package
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-slim
     if: github.repository_owner == 'ggml-org'
 
     steps:
diff --git a/src/llama-context.cpp b/src/llama-context.cpp
@@ -2173,13 +2173,6 @@ llm_graph_cb llama_context::graph_get_cb() const {
             ggml_set_name(cur, name);
         }
 
-        if (!cparams.offload_kqv) {
-            if (strcmp(name, "kqv_merged_cont") == 0) {
-                // all nodes between the KV store and the attention output are run on the CPU
-                ggml_backend_sched_set_tensor_backend(sched.get(), cur, backend_cpu);
-            }
-        }
-
         // norm may be automatically assigned to the backend of the previous layer, increasing data transfer between backends
         // FIXME: fix in ggml_backend_sched
         const bool full_offload = model.n_gpu_layers() > model.hparams.n_layer;
diff --git a/src/llama-graph.cpp b/src/llama-graph.cpp
@@ -1630,6 +1630,11 @@ ggml_tensor * llm_graph_context::build_attn_mha(
                                   hparams.attn_soft_cap ? hparams.f_attn_logit_softcapping : 0.0f);
         cb(cur, LLAMA_TENSOR_NAME_FATTN, il);
 
+        if (!cparams.offload_kqv) {
+            // all nodes between the KV store and the attention output are run on the CPU
+            ggml_backend_sched_set_tensor_backend(sched, cur, backend_cpu);
+        }
+
         ggml_flash_attn_ext_add_sinks(cur, sinks);
         ggml_flash_attn_ext_set_prec (cur, GGML_PREC_F32);
 
diff --git a/src/llama-model.cpp b/src/llama-model.cpp
@@ -1737,6 +1737,7 @@ void llama_model::load_hparams(llama_model_loader & ml) {
 
                 switch (hparams.n_layer) {
                     case 27: type = LLM_TYPE_16B; break;
+                    case 47: type = LLM_TYPE_30B_A3B; break;
                     case 60: type = LLM_TYPE_236B; break;
                     case 61: type = LLM_TYPE_671B; break;
                     default: type = LLM_TYPE_UNKNOWN;
