Update builder.sh

builder.sh

@@ -1,62 +1,45 @@
 #!/usr/bin/env bash
 set -euo pipefail
 
-echo "🚀 Builder (Apex + Q8) — runtime, GPU visível, cache persistente"
+echo "🚀 Builder (Apex + Q8) — roda em runtime com GPU visível"
 
-# ===== Persistência e caches =====
-# Prioriza /data (HF Spaces) e mantém compatibilidade com /app
-if [ -d /data ]; then
-  export HF_HOME="${HF_HOME:-/data/.cache/huggingface}"
-  export TORCH_HOME="${TORCH_HOME:-/data/.cache/torch}"
-else
-  export HF_HOME="${HF_HOME:-/app/.cache/huggingface}"
-  export TORCH_HOME="${TORCH_HOME:-/app/.cache/torch}"
-fi
+# ===== Config e diretórios =====
+export SELF_HF_REPO_ID="${SELF_HF_REPO_ID:-euIaxs22/Aduc-sdr}"   # Model repo no HF com wheels
+export HF_HOME="${HF_HOME:-/app/model_cache}"
 export HF_HUB_CACHE="${HF_HUB_CACHE:-$HF_HOME/hub}"
-mkdir -p "$HF_HOME" "$HF_HUB_CACHE" "$TORCH_HOME"
-
-# Symlink de compatibilidade (se scripts esperarem /app/.cache/huggingface)
-mkdir -p /app/.cache
-ln -sf "$HF_HOME" /app/.cache/huggingface
-
-# ===== Repositório de wheels no Hub =====
-export SELF_HF_REPO_ID="${SELF_HF_REPO_ID:-euIaxs22/Aduc-sdr}"
-
-# ===== Aceleração de transferência =====
+export TORCH_HOME="${TORCH_HOME:-$HF_HOME/torch}"
 export HF_HUB_ENABLE_HF_TRANSFER="${HF_HUB_ENABLE_HF_TRANSFER:-1}"
-export HF_HUB_DOWNLOAD_TIMEOUT="${HF_HUB_DOWNLOAD_TIMEOUT:-60}"
+export PATH="$HOME/.local/bin:$PATH"
 
-# ===== Diretórios de trabalho =====
-mkdir -p /app/wheels /app/cuda_cache /app/wheels/src
+mkdir -p /app/wheels /app/cuda_cache "$HF_HOME" "$TORCH_HOME" /app/wheels/src
 chmod -R 777 /app/wheels || true
 export CUDA_CACHE_PATH="/app/cuda_cache"
 
-# Licença (NVIDIA NGC) se presente
 if [ -f "/NGC-DL-CONTAINER-LICENSE" ]; then
   cp -f /NGC-DL-CONTAINER-LICENSE /app/wheels/NGC-DL-CONTAINER-LICENSE || true
 fi
 
-# ===== Dependências mínimas de build =====
+# ===== Dependências mínimas =====
 python -m pip install -v -U pip build setuptools wheel hatchling hatch-vcs scikit-build-core cmake ninja packaging "huggingface_hub[hf_transfer]" || true
 
 # ===== Tags de ambiente (Python/CUDA/Torch) =====
 PY_TAG="$(python -c 'import sys; print(f"cp{sys.version_info[0]}{sys.version_info[1]}")' 2>/dev/null || echo cp310)"
 TORCH_VER="$(python - <<'PY'
 try:
-  import torch, re
-  v = torch.__version__
-  print(re.sub(r'\+.*$', '', v))
+    import torch, re
+    v = torch.__version__
+    print(re.sub(r'\+.*$', '', v))
 except Exception:
-  print("unknown")
+    print("unknown")
 PY
 )"
 CU_TAG="$(python - <<'PY'
 try:
-  import torch
-  cu = getattr(torch.version, "cuda", None)
-  print("cu"+cu.replace(".","")) if cu else print("")
+    import torch
+    cu = getattr(torch.version, "cuda", None)
+    print("cu"+cu.replace(".","")) if cu else print("")
 except Exception:
-  print("")
+    print("")
 PY
 )"
 echo "[env] PY_TAG=${PY_TAG} TORCH_VER=${TORCH_VER} CU_TAG=${CU_TAG}"
@@ -65,11 +48,11 @@ echo "[env] PY_TAG=${PY_TAG} TORCH_VER=${TORCH_VER} CU_TAG=${CU_TAG}"
 check_apex() {
 python - <<'PY'
 try:
-  from apex.normalization import FusedLayerNorm, FusedRMSNorm
-  import importlib; importlib.import_module("fused_layer_norm_cuda")
-  ok = True
+    from apex.normalization import FusedLayerNorm, FusedRMSNorm
+    import importlib; importlib.import_module("fused_layer_norm_cuda")
+    ok = True
 except Exception:
-  ok = False
+    ok = False
 raise SystemExit(0 if ok else 1)
 PY
 }
@@ -82,10 +65,10 @@ raise SystemExit(0 if spec else 1)
 PY
 }
 
-# ===== Download de wheels do Hub =====
+# ===== Download do Hub =====
 install_from_hf () {
-  local PKG="$1" # 'apex' ou 'q8_kernels'
-  echo "[hub] Buscando wheel de ${PKG} em ${SELF_HF_REPO_ID} (py=${PY_TAG}, cu=${CU_TAG})"
+  local PKG="$1"    # 'apex' ou 'q8_kernels'
+  echo "[hub] Verificando wheel de ${PKG} no repositório ${SELF_HF_REPO_ID}"
   python - "$PKG" "$PY_TAG" "$CU_TAG" <<'PY' || exit 0
 import os, sys
 from huggingface_hub import HfApi, hf_hub_download, HfFolder
@@ -93,16 +76,16 @@ from huggingface_hub import HfApi, hf_hub_download, HfFolder
 pkg, py_tag, cu_tag = sys.argv[1], sys.argv[2], sys.argv[3]
 repo = os.environ.get("SELF_HF_REPO_ID","euIaxs22/Aduc-sdr")
 api = HfApi(token=os.getenv("HF_TOKEN") or HfFolder.get_token())
-
 try:
-  files = api.list_repo_files(repo_id=repo, repo_type="model")
+    files = api.list_repo_files(repo_id=repo, repo_type="model")
 except Exception:
-  raise SystemExit(0)
+    raise SystemExit(0)
 
 cands = [f for f in files if f.endswith(".whl") and f.rsplit("/",1)[-1].startswith(pkg+"-") and py_tag in f]
-pref = [f for f in cands if cu_tag and cu_tag in f] or cands
+pref  = [f for f in cands if cu_tag and cu_tag in f] or cands
 if not pref:
-  raise SystemExit(0)
+    raise SystemExit(0)
+
 target = sorted(pref, reverse=True)[0]
 print(target)
 path = hf_hub_download(repo_id=repo, filename=target, repo_type="model", local_dir="/app/wheels")
@@ -113,7 +96,7 @@ PY
 # ===== Builders =====
 build_apex () {
   local SRC="/app/wheels/src/apex"
-  echo "[build] Fonte Apex em ${SRC}"
+  echo "[build] Preparando fonte Apex em ${SRC}"
   if [ -d "$SRC/.git" ]; then
     git -C "$SRC" fetch --all -p || true
     git -C "$SRC" reset --hard HEAD || true
@@ -122,18 +105,16 @@ build_apex () {
     rm -rf "$SRC"
     git clone --depth 1 https://github.com/NVIDIA/apex "$SRC"
   fi
-
   echo "[build] Compilando Apex -> wheel"
   export APEX_CPP_EXT=1 APEX_CUDA_EXT=1 APEX_ALL_CONTRIB_EXT=0
-  python -m pip wheel --no-build-isolation --no-deps "$SRC" -w /app/wheels || true
-
+  python -m pip wheel -v --no-build-isolation --no-deps "$SRC" -w /app/wheels || true
   local W="$(ls -t /app/wheels/apex-*.whl 2>/dev/null | head -n1 || true)"
   if [ -n "${W}" ]; then
-    python -m pip install -U --no-deps "${W}" || true
-    echo "[build] Apex instalado da wheel: ${W}"
+    python -m pip install -v -U --no-deps "${W}" || true
+    echo "[build] Apex instalado da wheel recém-compilada: ${W}"
   else
-    echo "[build] Nenhuma wheel Apex gerada; instalando do source"
-    python -m pip install --no-build-isolation "$SRC" || true
+    echo "[build] Nenhuma wheel Apex gerada; instalando do source (pode falhar)"
+    python -m pip install -v --no-build-isolation "$SRC" || true
   fi
 }
 
@@ -145,27 +126,25 @@ build_q8 () {
   git clone --filter=blob:none "$Q8_REPO" "$SRC"
   git -C "$SRC" checkout "$Q8_COMMIT"
   git -C "$SRC" submodule update --init --recursive
-
   echo "[build] Compilando Q8 Kernels -> wheel"
-  python -m pip wheel --no-build-isolation "$SRC" -w /app/wheels || true
-
+  python -m pip wheel -v --no-build-isolation "$SRC" -w /app/wheels || true
   local W="$(ls -t /app/wheels/q8_kernels-*.whl 2>/dev/null | head -n1 || true)"
   if [ -n "${W}" ]; then
-    python -m pip install  -U --no-deps "${W}" || true
-    echo "[build] Q8 instalado da wheel: ${W}"
+    python -m pip install -v -U --no-deps "${W}" || true
+    echo "[build] Q8 instalado da wheel recém-compilada: ${W}"
   else
-    echo "[build] Nenhuma wheel q8_kernels gerada; instalando do source"
-    python -m pip install  --no-build-isolation "$SRC" || true
+    echo "[build] Nenhuma wheel q8_kernels gerada; instalando do source (pode falhar)"
+    python -m pip install -v --no-build-isolation "$SRC" || true
   fi
 }
 
 # ===== Pipeline genérico =====
 ensure_pkg () {
-  local PKG="$1"       # apex | q8_kernels
-  local CHECK_FN="$2"  # check_apex | check_q8
-  local BUILD_FN="$3"  # build_apex | build_q8
-  echo "[flow] === ${PKG} ==="
+  local PKG="$1"      # apex | q8_kernels
+  local CHECK_FN="$2" # check_apex | check_q8
+  local BUILD_FN="$3" # build_apex | build_q8
 
+  echo "[flow] === ${PKG} ==="
   if ${CHECK_FN}; then
     echo "[flow] ${PKG}: já instalado (import OK)"
     return 0
@@ -176,7 +155,7 @@ ensure_pkg () {
   if [ -n "${HF_OUT:-}" ]; then
     WHEEL_PATH="$(printf "%s\n" "${HF_OUT}" | tail -n1)"
     echo "[hub] Baixado: ${WHEEL_PATH}"
-    python -m pip install -U --no-build-isolation "${WHEEL_PATH}" || true
+    python -m pip install -v -U --no-build-isolation "${WHEEL_PATH}" || true
     if ${CHECK_FN}; then
       echo "[flow] ${PKG}: sucesso via Hub (${WHEEL_PATH})"
       return 0
@@ -194,102 +173,33 @@ ensure_pkg () {
     return 0
   fi
 
-  echo "[flow] ${PKG}: falhou após build; seguindo adiante"
+  echo "[flow] ${PKG}: falhou após build; registrando logs e seguindo"
   return 1
 }
 
-
-# ===== layer_norm (wheel separada) =====
-check_layernorm () { python - <<'PY'
-import importlib, sys
-try:
-    importlib.import_module("layer_norm_cuda")
-    sys.exit(0)
-except Exception:
-    sys.exit(1)
-PY
-}
-install_layernorm_from_hf () {
-  log "[hub] tentando wheel 'layer_norm' no ${SELF_HF_REPO_ID} (py=${PY_TAG} cu=${CU_TAG} torch=${TORCH_VER})"
-  python - <<'PY' 2>/dev/null || exit 0
-from huggingface_hub import HfApi, HfFolder
-import os, re, sys
-repo = os.environ.get("SELF_HF_REPO_ID")
-py = os.environ.get("PY_TAG"); cu = os.environ.get("CU_TAG"); torchv = os.environ.get("TORCH_VER")
-api = HfApi(token=os.getenv("HF_TOKEN") or HfFolder.get_token())
-try:
-    files = api.list_repo_files(repo_id=repo, repo_type="model")
-except Exception:
-    sys.exit(0)
-p1 = re.compile(rf"(?:^|/)layer[_-]?norm-.*{re.escape(py)}.*{re.escape(cu)}.*{re.escape(torchv)}.*\.whl$")
-p2 = re.compile(rf"(?:^|/)(?:flash[_-]?attn[_-])?layer[_-]?norm-.*{re.escape(py)}.*{re.escape(cu)}.*\.whl$")
-cands = [f for f in files if f.endswith(".whl") and (p1.search(f) or p2.search(f))]
-if not cands: sys.exit(0)
-cands.sort(reverse=True)
-path = api.hf_hub_download(repo_id=repo, filename=cands[0], repo_type="model", local_dir="/app/wheels")
-print(path)
-PY
-}
-build_layernorm () {
-  local SRC="/app/wheels/src/flash-attention"
-  if [ ! -d "$SRC/.git" ]; then
-    rm -rf "$SRC"; git clone --filter=blob:none "$FLASHATTN_REPO" "$SRC"
-  fi
-  git -C "$SRC" fetch --all -p || true
-  git -C "$SRC" checkout "$FLASHATTN_REF" || true
-  git -C "$SRC" submodule update --init --recursive || true
-  export MAX_JOBS="${MAX_JOBS:-4}"
-  log "[build] layer_norm — MAX_JOBS=$MAX_JOBS"
-  (cd "$SRC/csrc/layer_norm" && python -m pip wheel --no-build-isolation --no-deps . -w /app/wheels)
-  local W="$(ls -t /app/wheels/*layer*norm*-*.whl 2>/dev/null | head -n1 || true)"
-  if [ -n "$W" ]; then
-    python -m pip install -U --no-deps --no-build-isolation "$W" || true
-    log "[build] layer_norm instalado da wheel: $W"
-  else
-    log "[build] layer_norm: wheel não encontrada; instalando do source"
-    (cd "$SRC/csrc/layer_norm" && MAX_JOBS="$MAX_JOBS" pip install . --no-build-isolation) || true
-  fi
-}
-ensure_layernorm () {
-  log "[flow] === layer_norm (FlashAttention) ==="
-  if check_layernorm; then log "[flow] layer_norm já instalado (import OK)"; return 0; fi
-  local HF_LN="$(install_layernorm_from_hf || true)"
-  if [ -n "${HF_LN:-}" ]; then
-    python -m pip install -U --no-deps --no-build-isolation "$HF_LN" || true
-    if check_layernorm; then log "[flow] layer_norm via Hub OK"; return 0; fi
-  fi
-  log "[flow] compilando layer_norm (fallback)"
-  build_layernorm
-  if check_layernorm; then log "[flow] layer_norm build OK"; return 0; fi
-  log "[flow] layer_norm falhou"; return 1
-}
-
-
-
 # ===== Execução: Apex e Q8 =====
-
 ensure_pkg "apex" check_apex build_apex || true
-ensure_pkg "q8_kernels" check_q8 build_q8 || true
-ensure_layernorm || true
+#ensure_pkg "q8_kernels" check_q8 build_q8 || true
+
 
-# ===== Upload das wheels geradas (opcional) =====
 python - <<'PY'
 import os
 from huggingface_hub import HfApi, HfFolder
 repo=os.environ.get("SELF_HF_REPO_ID","euIaxs22/Aduc-sdr")
 token=os.getenv("HF_TOKEN") or HfFolder.get_token()
 if not token:
-  raise SystemExit("HF_TOKEN ausente; upload desabilitado")
+    raise SystemExit("HF_TOKEN ausente; upload desabilitado")
 api=HfApi(token=token)
 api.upload_folder(
-  folder_path="/app/wheels",
-  repo_id=repo,
-  repo_type="model",
-  allow_patterns=["*.whl","NGC-DL-CONTAINER-LICENSE"],
-  ignore_patterns=["**/src/**","**/*.log","**/logs/**",".git/**"],
+    folder_path="/app/wheels",
+    repo_id=repo,
+    repo_type="model",
+    allow_patterns=["*.whl","NGC-DL-CONTAINER-LICENSE"],
+    ignore_patterns=["**/src/**","**/*.log","**/logs/**",".git/**"],
 )
 print("Upload concluído (wheels + licença).")
 PY
 
+
 chmod -R 777 /app/wheels || true
 echo "✅ Builder finalizado."