data: sub 694 (683分) 日志 + resolve merge

probe_model_shapes.sh

@@ -0,0 +1,48 @@
+#!/bin/bash
+set -e
+
+echo "=== 模型权重实际shape ==="
+python3 -c "
+import torch, os, json
+# 读config.json
+cfg_path = '/model/config.json'
+if os.path.exists(cfg_path):
+    with open(cfg_path) as f:
+        cfg = json.load(f)
+    print('Model config:')
+    for k in ['hidden_size', 'intermediate_size', 'num_attention_heads', 
+              'num_key_value_heads', 'num_hidden_layers', 'num_experts',
+              'num_experts_per_tok', 'moe_intermediate_size', 'vocab_size',
+              'max_position_embeddings']:
+        print(f'  {k}: {cfg.get(k, \"N/A\")}')
+else:
+    print(f'{cfg_path} not found')
+    # 搜索
+    import glob
+    for p in glob.glob('/model/**/config.json', recursive=True):
+        print(f'  found: {p}')
+"
+
+echo ""
+echo "=== safetensor权重shape（第一个shard）==="
+python3 -c "
+from safetensors import safe_open
+import glob, os
+shards = sorted(glob.glob('/model/model*.safetensors'))
+if not shards:
+    shards = sorted(glob.glob('/model/*.safetensors'))
+if shards:
+    print(f'Found {len(shards)} shards, reading first: {shards[0]}')
+    with safe_open(shards[0], framework='pt') as f:
+        for key in sorted(f.keys()):
+            if 'experts' in key and ('w1' in key or 'w2' in key or 'w13' in key):
+                print(f'  {key}: {f.get_tensor(key).shape}')
+                break  # 只看一个就够了
+        # 也看gate
+        for key in sorted(f.keys()):
+            if 'gate' in key and 'weight' in key:
+                print(f'  {key}: {f.get_tensor(key).shape}')
+                break
+else:
+    print('No safetensor shards found')
+" 2>&1 || echo "safetensors not available"

probe_so_output.txt

@@ -10,9 +10,9 @@
 === 2. Python import路径 ===
 /usr/local/corex/lib64/python3/dist-packages/torch/cuda/__init__.py:51: FutureWarning: The pynvml package is deprecated. Please install nvidia-ml-py instead. If you did not install pynvml directly, please report this to the maintainers of the package that installed pynvml for you.
   import pynvml  # type: ignore[import]
-INFO 08-15 14:53:14 importing.py:10] Triton not installed; certain GPU-related functions will not be available.
-2026-08-15 14:53:16.346469: I tensorflow/core/util/port.cc:110] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.
-2026-08-15 14:53:16.398804: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
+INFO 08-15 14:48:59 importing.py:10] Triton not installed; certain GPU-related functions will not be available.
+2026-08-15 14:49:01.632894: I tensorflow/core/util/port.cc:110] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.
+2026-08-15 14:49:01.686627: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
 To enable the following instructions: SSE3 SSE4.1 SSE4.2 AVX AVX2 AVX512F AVX512_VNNI AVX512_BF16 AVX_VNNI AMX_TILE AMX_INT8 AMX_BF16 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
 WARNING:tensorflow:Deprecation warnings have been disabled. Set TF_ENABLE_DEPRECATION_WARNINGS=1 to re-enable them.
 vllm.__file__: /home/dylan/0814/project_6/vllm/__init__.py
@@ -37,9 +37,9 @@ vllm dir: /home/dylan/0814/project_6/vllm
 === 3. 逐个import corex_moe测试 ===
 /usr/local/corex/lib64/python3/dist-packages/torch/cuda/__init__.py:51: FutureWarning: The pynvml package is deprecated. Please install nvidia-ml-py instead. If you did not install pynvml directly, please report this to the maintainers of the package that installed pynvml for you.
   import pynvml  # type: ignore[import]
-INFO 08-15 14:53:25 importing.py:10] Triton not installed; certain GPU-related functions will not be available.
-2026-08-15 14:53:27.639666: I tensorflow/core/util/port.cc:110] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.
-2026-08-15 14:53:27.692162: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
+INFO 08-15 14:49:11 importing.py:10] Triton not installed; certain GPU-related functions will not be available.
+2026-08-15 14:49:13.043593: I tensorflow/core/util/port.cc:110] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.
+2026-08-15 14:49:13.095797: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
 To enable the following instructions: SSE3 SSE4.1 SSE4.2 AVX AVX2 AVX512F AVX512_VNNI AVX512_BF16 AVX_VNNI AMX_TILE AMX_INT8 AMX_BF16 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
 WARNING:tensorflow:Deprecation warnings have been disabled. Set TF_ENABLE_DEPRECATION_WARNINGS=1 to re-enable them.
   ✓ from vllm import corex_moe_topk_softmax → ['moe_topk_softmax']
@@ -61,9 +61,9 @@ WARNING:tensorflow:Deprecation warnings have been disabled. Set TF_ENABLE_DEPREC
 === 4. ix_unified_bridge import测试 ===
 /usr/local/corex/lib64/python3/dist-packages/torch/cuda/__init__.py:51: FutureWarning: The pynvml package is deprecated. Please install nvidia-ml-py instead. If you did not install pynvml directly, please report this to the maintainers of the package that installed pynvml for you.
   import pynvml  # type: ignore[import]
-INFO 08-15 14:53:37 importing.py:10] Triton not installed; certain GPU-related functions will not be available.
-2026-08-15 14:53:38.942008: I tensorflow/core/util/port.cc:110] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.
-2026-08-15 14:53:38.994486: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
+INFO 08-15 14:49:22 importing.py:10] Triton not installed; certain GPU-related functions will not be available.
+2026-08-15 14:49:24.345485: I tensorflow/core/util/port.cc:110] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.
+2026-08-15 14:49:24.397567: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
 To enable the following instructions: SSE3 SSE4.1 SSE4.2 AVX AVX2 AVX512F AVX512_VNNI AVX512_BF16 AVX_VNNI AMX_TILE AMX_INT8 AMX_BF16 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
 WARNING:tensorflow:Deprecation warnings have been disabled. Set TF_ENABLE_DEPRECATION_WARNINGS=1 to re-enable them.
   ✗ ix_unified_bridge: cannot import name 'ix_unified_bridge' from 'vllm' (/home/dylan/0814/project_6/vllm/__init__.py)
@@ -71,9 +71,9 @@ WARNING:tensorflow:Deprecation warnings have been disabled. Set TF_ENABLE_DEPREC
 === 5. 我们的qwen3_5.py里各flag的实际值 ===
 /usr/local/corex/lib/python3/dist-packages/torch/cuda/__init__.py:51: FutureWarning: The pynvml package is deprecated. Please install nvidia-ml-py instead. If you did not install pynvml directly, please report this to the maintainers of the package that installed pynvml for you.
   import pynvml  # type: ignore[import]
-INFO 08-15 14:53:48 importing.py:10] Triton not installed; certain GPU-related functions will not be available.
-2026-08-15 14:53:50.143129: I tensorflow/core/util/port.cc:110] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.
-2026-08-15 14:53:50.195319: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
+INFO 08-15 14:49:33 importing.py:10] Triton not installed; certain GPU-related functions will not be available.
+2026-08-15 14:49:35.533303: I tensorflow/core/util/port.cc:110] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.
+2026-08-15 14:49:35.585311: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
 To enable the following instructions: SSE3 SSE4.1 SSE4.2 AVX AVX2 AVX512F AVX512_VNNI AVX512_BF16 AVX_VNNI AMX_TILE AMX_INT8 AMX_BF16 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
 WARNING:tensorflow:Deprecation warnings have been disabled. Set TF_ENABLE_DEPRECATION_WARNINGS=1 to re-enable them.
   ✗ _USE_COREX_BATCHED_GEMM = False

qwen3_6_scripts/probe_all_symbols.sh

@@ -0,0 +1,120 @@
+#!/bin/bash
+# probe_all_symbols.sh — Dump ALL exported symbols from every relevant .so
+# No grep filter — save full lists, then we search offline
+
+OUTDIR="cat_files/symbol_dumps"
+mkdir -p "$OUTDIR"
+
+echo "=== 1. ALL ixformer .so files ==="
+find /usr/local/corex -name "*ixformer*" -name "*.so" 2>/dev/null | sort | tee "$OUTDIR/ixformer_so_list.txt"
+echo ""
+
+echo "=== 2. Dump each ixformer .so symbols ==="
+while read so; do
+    base=$(basename "$so" | sed 's/[^a-zA-Z0-9._-]/_/g')
+    count=$(nm -D "$so" 2>/dev/null | grep " T " | wc -l)
+    echo "  $so → $base ($count T symbols)"
+    nm -D "$so" 2>/dev/null | grep " T " > "$OUTDIR/sym_${base}.txt"
+done < "$OUTDIR/ixformer_so_list.txt"
+echo ""
+
+echo "=== 3. libixformer.so full T symbols ==="
+if [ -f /usr/local/corex/lib64/libixformer.so ]; then
+    nm -D /usr/local/corex/lib64/libixformer.so 2>/dev/null | grep " T " > "$OUTDIR/sym_libixformer.txt"
+    wc -l "$OUTDIR/sym_libixformer.txt"
+    # Also search for ANY moe/expert/gemm/fused related
+    echo "  grep moe/expert/gemm/fused/group/batch:"
+    grep -i "moe\|expert\|gemm\|fused\|group\|batch\|topk\|gating\|route" "$OUTDIR/sym_libixformer.txt" | head -30
+fi
+echo ""
+
+echo "=== 4. libcuinfer.so full T symbols ==="
+if [ -f /usr/local/corex/lib64/libcuinfer.so ]; then
+    nm -D /usr/local/corex/lib64/libcuinfer.so 2>/dev/null | grep " T " > "$OUTDIR/sym_libcuinfer.txt"
+    wc -l "$OUTDIR/sym_libcuinfer.txt"
+    echo "  grep moe/expert/gemm/fused/group/batch:"
+    grep -i "moe\|expert\|gemm\|fused\|group\|batch\|topk\|gating\|route" "$OUTDIR/sym_libcuinfer.txt" | head -30
+fi
+echo ""
+
+echo "=== 5. _ixformer_torch .so full T symbols ==="
+TORCH_SO=$(find /usr/local/corex -name "_ixformer_torch*.so" 2>/dev/null | head -1)
+if [ -n "$TORCH_SO" ]; then
+    nm -D "$TORCH_SO" 2>/dev/null | grep " T " > "$OUTDIR/sym_ixformer_torch.txt"
+    wc -l "$OUTDIR/sym_ixformer_torch.txt"
+    echo "  grep moe/expert/gemm/fused/group/batch:"
+    grep -i "moe\|expert\|gemm\|fused\|group\|batch\|topk\|gating\|route" "$OUTDIR/sym_ixformer_torch.txt" | head -30
+fi
+echo ""
+
+echo "=== 6. _C .so (ixformer python binding) full T symbols ==="
+C_SO=$(find /usr/local/corex -path "*ixformer*" -name "_C*.so" 2>/dev/null | head -1)
+if [ -n "$C_SO" ]; then
+    nm -D "$C_SO" 2>/dev/null | grep " T " > "$OUTDIR/sym_ixformer_C.txt"
+    wc -l "$OUTDIR/sym_ixformer_C.txt"
+    echo "  grep moe/expert/gemm/fused/group/batch:"
+    grep -i "moe\|expert\|gemm\|fused\|group\|batch\|topk\|gating\|route" "$OUTDIR/sym_ixformer_C.txt" | head -30
+fi
+echo ""
+
+echo "=== 7. ALL .so in ixformer package dir ==="
+IXDIR=$(python3 -c "import ixformer, os; print(os.path.dirname(ixformer.__file__))" 2>/dev/null)
+if [ -n "$IXDIR" ]; then
+    echo "ixformer dir: $IXDIR"
+    find "$IXDIR" -name "*.so" | while read so; do
+        base=$(basename "$so")
+        count=$(nm -D "$so" 2>/dev/null | grep " T " | wc -l)
+        echo "  $base: $count T symbols"
+        nm -D "$so" 2>/dev/null | grep " T " > "$OUTDIR/sym_ixpkg_${base}.txt"
+        # Quick search
+        hits=$(grep -ic "moe\|expert\|gemm\|fused\|group\|batch\|topk" "$OUTDIR/sym_ixpkg_${base}.txt")
+        if [ "$hits" -gt 0 ]; then
+            echo "    *** HIT: $hits MoE/GEMM related symbols:"
+            grep -i "moe\|expert\|gemm\|fused\|group\|batch\|topk" "$OUTDIR/sym_ixpkg_${base}.txt"
+        fi
+    done
+fi
+echo ""
+
+echo "=== 8. ixformer Python API — list ALL callable functions ==="
+python3 << 'PY'
+import ixformer
+import inspect
+
+# List all attributes
+for name in sorted(dir(ixformer)):
+    if name.startswith('_'):
+        continue
+    obj = getattr(ixformer, name)
+    if callable(obj):
+        try:
+            sig = inspect.signature(obj)
+            print(f"  ixformer.{name}{sig}")
+        except (ValueError, TypeError):
+            print(f"  ixformer.{name} (no signature)")
+    elif hasattr(obj, '__module__'):
+        print(f"  ixformer.{name} = {type(obj).__name__}")
+
+# Check submodules
+print("\n  --- submodules ---")
+for name in sorted(dir(ixformer)):
+    obj = getattr(ixformer, name)
+    if inspect.ismodule(obj) and not name.startswith('_'):
+        print(f"  ixformer.{name}:")
+        for sub in sorted(dir(obj)):
+            if sub.startswith('_'):
+                continue
+            subobj = getattr(obj, sub)
+            if callable(subobj):
+                try:
+                    sig = inspect.signature(subobj)
+                    print(f"    .{sub}{sig}")
+                except:
+                    print(f"    .{sub} (no sig)")
+PY
+
+echo ""
+echo "=== Files saved to $OUTDIR ==="
+ls -lh "$OUTDIR/"
+echo ""
+echo "git add cat_files/symbol_dumps/ && git commit -m 'data: full symbol dumps' && git push"