some fixes to Dockerfile

Dockerfile

@@ -58,6 +58,13 @@ RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
 WORKDIR /workspace
 RUN python3 -m venv /workspace/py-venv
 ENV PATH="/workspace/py-venv/bin:${PATH}"
+
+# llama.cpp requirements
+COPY --from=builder --chmod=0644 /workspace/requirements.txt /workspace/requirements-llama.txt
+COPY --from=builder /workspace/requirements /workspace/requirements
+RUN --mount=type=cache,target=/root/.cache/pip,sharing=locked pip3 install -r /workspace/requirements-llama.txt
+
+# Upgrade torch after llama's downgrade
 RUN --mount=type=cache,target=/root/.cache/pip,sharing=locked pip3 install torch torchvision torchaudio --upgrade --index-url \
     https://download.pytorch.org/whl/cu132/
 
@@ -69,8 +76,8 @@ RUN git clone --depth=1 https://github.com/silveroxides/convert_to_quant && \
 RUN --mount=type=cache,target=/root/.cache/pip,sharing=locked pip3 install heretic-llm
 
 # Install zefie's nvfp4 quant requirements
-COPY --chmod=0644 requirements.txt /workspace/requirements.txt
+COPY --chmod=0644 requirements.txt /workspace/requirements-nvfp4.txt
-RUN --mount=type=cache,target=/root/.cache/pip,sharing=locked pip3 install -r /workspace/requirements.txt
+RUN --mount=type=cache,target=/root/.cache/pip,sharing=locked pip3 install -r /workspace/requirements-nvfp4.txt
 
 # ---------- Main image