tocmo0nlord/axolotl
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

Update SETUP_MIAAI.md: pre-training checklist, Ollama stop/start, verify script, corrected training time

Browse Source
This commit is contained in:
tocmo0nlord
2026-05-13 21:19:15 +00:00
parent 981a13e110
commit c7c4885369
1 changed files with 101 additions and 66 deletions
Download Patch File Download Diff File Expand all files Collapse all files

167
SETUP_MIAAI.md
View File

@@ -4,11 +4,57 @@
- GPU: NVIDIA RTX 5080 (16GB VRAM, sm_120 / Blackwell) - GPU: NVIDIA RTX 5080 (16GB VRAM, sm_120 / Blackwell)
- Driver: 580.126.09 — max CUDA 13.0 shown by nvidia-smi, but nvcc from conda is 13.2 - Driver: 580.126.09 — max CUDA 13.0 shown by nvidia-smi, but nvcc from conda is 13.2
- OS: Ubuntu 25.10 (Python 3.13 system — do NOT use system Python for ML) - OS: Ubuntu 25.10 (Python 3.13 system — do NOT use system Python for ML)
- Axolotl branch: `activeblue/main` - Axolotl repo: `/home/tocmo0nlord/axolotl` (branch: `activeblue/main`)
- Conda env: `axolotl` at `/opt/miniconda3/envs/axolotl`
--- ---
## One-time Setup ## Pre-Training Checklist (every time)
Before starting a training run, verify these:
```bash
# 1. Stop Ollama — if a request hits it mid-training it will compete for VRAM
sudo systemctl stop ollama
# 2. Activate conda env
export PATH="/opt/miniconda3/bin:$PATH"
conda activate axolotl
# 3. Set env vars
export CUDA_HOME=$CONDA_PREFIX
export PATH=$CUDA_HOME/bin:$PATH
export PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True
# 4. Confirm GPU is clear (should show no processes)
nvidia-smi --query-compute-apps=pid,process_name,used_memory --format=csv
# 5. Go to axolotl directory
cd /home/tocmo0nlord/axolotl
```
## Run Training
```bash
axolotl train ~/human_chat_qlora.yml
```
## After Training
```bash
# Restart Ollama
sudo systemctl start ollama
# Test the adapter
axolotl inference ~/human_chat_qlora.yml \
--lora-model-dir ~/outputs/llama31-8b-humanchat \
--prompter chat
# (Optional) Merge adapter into base model
axolotl merge-lora ~/human_chat_qlora.yml
```
---
## One-time Setup (fresh machine only)
### 1. Install Miniconda ### 1. Install Miniconda
```bash ```bash
@@ -24,14 +70,13 @@ conda create -n axolotl python=3.11 -y
conda activate axolotl conda activate axolotl
``` ```
### 3. Clone and sync repo with upstream ### 3. Clone axolotl repo
```bash ```bash
git clone https://git.activeblue.net/tocmo0nlord/axolotl.git git clone https://git.activeblue.net/tocmo0nlord/axolotl.git /home/tocmo0nlord/axolotl
cd axolotl cd /home/tocmo0nlord/axolotl
git remote add upstream https://github.com/axolotl-ai-cloud/axolotl.git git remote add upstream https://github.com/axolotl-ai-cloud/axolotl.git
git fetch upstream git fetch upstream
git rebase upstream/main # keeps activeblue patches on top git rebase upstream/main # keeps activeblue patches on top
git push origin activeblue/main --force-with-lease
``` ```
### 4. Install CUDA toolkit (needed to compile flash-attn and bitsandbytes) ### 4. Install CUDA toolkit (needed to compile flash-attn and bitsandbytes)
@@ -45,55 +90,53 @@ export PATH=$CUDA_HOME/bin:$PATH
> This is fine — use cu132 everywhere to match. > This is fine — use cu132 everywhere to match.
### 5. Install PyTorch — use cu132 (matches nvcc from conda) ### 5. Install PyTorch — use cu132 (matches nvcc from conda)
> NOTE: torchaudio has no cu132 wheel — skip it, not needed for LLM training
```bash ```bash
# torchaudio has no cu132 wheel — skip it, not needed for LLM training
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu132 pip install torch torchvision --index-url https://download.pytorch.org/whl/cu132
python -c "import torch; print('CUDA:', torch.version.cuda); print('GPU:', torch.cuda.get_device_name(0))" python -c "import torch; print('CUDA:', torch.version.cuda); print('GPU:', torch.cuda.get_device_name(0))"
``` ```
### 6. Install Axolotl ### 6. Install Axolotl
```bash ```bash
cd /home/tocmo0nlord/axolotl
pip install -e "." pip install -e "."
``` ```
> **flash-attn compiles CUDA kernels from source — takes 15–25 min on 10 cores of i7-14700K.** ### 7. Install flash-attn
> Always set `MAX_JOBS` to the number of available CPU cores: > Compiles CUDA kernels from source — takes 15–25 min on 10 cores of i7-14700K.
```bash ```bash
MAX_JOBS=10 pip install flash-attn --no-build-isolation MAX_JOBS=10 pip install flash-attn --no-build-isolation
``` ```
### 7. Compile bitsandbytes from source for sm_120 (RTX 5080 / Blackwell) ### 8. Compile bitsandbytes from source for sm_120 (RTX 5080 / Blackwell)
The prebuilt bitsandbytes wheels do not include sm_120 support and CUDA 13.2 dropped sm_50–53. Prebuilt wheels do not include sm_120. CUDA 13.2 also dropped sm_50–53.
You must compile from source with a patched CMakeLists.txt. Must compile from source with a patched CMakeLists.txt.
```bash ```bash
# Clone bitsandbytes v0.49.1 # Clone bitsandbytes v0.49.1
git clone --branch v0.49.1 --depth 1 https://github.com/bitsandbytes-foundation/bitsandbytes.git /tmp/bnb_0491 git clone --branch v0.49.1 --depth 1 https://github.com/bitsandbytes-foundation/bitsandbytes.git /tmp/bnb_0491
cd /tmp/bnb_0491
# Patch CMakeLists.txt: override arch list to sm_120 only, just before the foreach loop # Patch CMakeLists.txt: insert sm_120 override before the foreach loop
# (cmake >= 3.23.0 skips the manual arch block and uses its own built-in list which lacks sm_120) # (cmake >= 3.23.0 uses its own built-in arch list which does not include sm_120)
sed -i '/ foreach(capability \${CMAKE_CUDA_ARCHITECTURES_ALL})/i\ # RTX 5080 sm_120 patch: override before capability list is built\n set(CMAKE_CUDA_ARCHITECTURES_ALL 120)' CMakeLists.txt sed -i '/ foreach(capability \${CMAKE_CUDA_ARCHITECTURES_ALL})/i\ # RTX 5080 sm_120 patch\n set(CMAKE_CUDA_ARCHITECTURES_ALL 120)' /tmp/bnb_0491/CMakeLists.txt
# Verify the patch landed at the right line # Verify patch landed correctly (should show the set() line immediately before foreach)
grep -n "ARCHITECTURES_ALL\|foreach" CMakeLists.txt | tail -5 grep -n "ARCHITECTURES_ALL\|foreach" /tmp/bnb_0491/CMakeLists.txt | tail -5
# Should show: set(CMAKE_CUDA_ARCHITECTURES_ALL 120) immediately before the foreach line
# Configure — must point cmake at conda's nvcc # Configure
cmake \ cmake \
-DCMAKE_CUDA_COMPILER=/opt/miniconda3/envs/axolotl/bin/nvcc \ -DCMAKE_CUDA_COMPILER=/opt/miniconda3/envs/axolotl/bin/nvcc \
-DCOMPUTE_BACKEND=cuda \ -DCOMPUTE_BACKEND=cuda \
-S /tmp/bnb_0491 \ -S /tmp/bnb_0491 \
-B /tmp/bnb_0491/build 2>&1 | grep -E "(Capabilit|CUDA Ver|Error)" -B /tmp/bnb_0491/build 2>&1 | grep -E "(Capabilit|CUDA Ver|Error)"
# Expected: "CUDA Capabilities Selected: 120" # Must show: CUDA Capabilities Selected: 120
# Build (j10 uses 10 cores — adjust to your CPU) # Build
cmake --build /tmp/bnb_0491/build -j10 cmake --build /tmp/bnb_0491/build -j10
# Install into conda site-packages # Install into conda site-packages
SITE_PKG=/opt/miniconda3/envs/axolotl/lib/python3.11/site-packages cp -r /tmp/bnb_0491/bitsandbytes /opt/miniconda3/envs/axolotl/lib/python3.11/site-packages/
cp -r /tmp/bnb_0491/bitsandbytes "$SITE_PKG/"
# Verify # Verify
python3 -c " python3 -c "
@@ -104,61 +147,52 @@ print('bitsandbytes CUDA OK:', l(x).shape)
" "
``` ```
--- ### 9. HuggingFace login (meta-llama is gated)
## Every Session (after first-time setup)
```bash ```bash
export PATH="/opt/miniconda3/bin:$PATH" huggingface-cli login
conda activate axolotl # Paste your HF token when prompted
export CUDA_HOME=$CONDA_PREFIX ```
export PATH=$CUDA_HOME/bin:$PATH
export PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True ### 10. Verify everything is working
cd /home/tocmo0nlord/axolotl ```bash
python3 -c "
import torch, bitsandbytes as bnb, flash_attn, transformers, axolotl
print('torch:', torch.__version__, '| CUDA:', torch.version.cuda)
print('bitsandbytes:', bnb.__version__)
print('flash_attn:', flash_attn.__version__)
print('transformers:', transformers.__version__)
print('GPU:', torch.cuda.get_device_name(0))
print('VRAM:', round(torch.cuda.get_device_properties(0).total_memory/1e9, 1), 'GB')
"
``` ```
--- ---
## Training Config — human_chat_qlora.yml ## Training Config — human_chat_qlora.yml
Key settings that work on RTX 5080 (16GB): Key settings tuned for RTX 5080 (16GB):
| Setting | Value | Notes | | Setting | Value | Notes |
|---|---|---| |---|---|---|
| `sequence_len` | `2048` | 4096 causes OOM during loss computation (logits x 128k vocab) | | `sequence_len` | `2048` | 4096 OOMs during loss computation (logits x 128k vocab) |
| `micro_batch_size` | `1` | Keep low; effective batch = micro x grad_accum | | `micro_batch_size` | `1` | Effective batch = micro x grad_accum = 8 |
| `gradient_accumulation_steps` | `8` | Effective batch = 8 | | `gradient_accumulation_steps` | `8` | Keeps effective batch at 8 |
| `adapter` | `qlora` | QLoRA 4-bit via bitsandbytes | | `adapter` | `qlora` | 4-bit via bitsandbytes compiled from source |
| `attn_implementation` | `flash_attention_2` | Not the deprecated `flash_attention: true` | | `attn_implementation` | `flash_attention_2` | Not the deprecated `flash_attention: true` |
| `type` (datasets) | `chat_template` | Not the deprecated `sharegpt` | | `type` (datasets) | `chat_template` | Not the deprecated `sharegpt` |
Dataset fields for SlimOrca / OpenHermes-2.5 (sharegpt-format with different field names): Expected training metrics (RTX 5080, ~65k samples, 2 epochs):
```yaml - VRAM: ~10–11 GB active, ~11 GB allocated
datasets: - Training duration: ~3.5 hours
- path: Open-Orca/SlimOrca - Initial eval loss: ~0.81, perplexity ~2.25
type: chat_template - Final loss target: ~0.55–0.60
field_messages: conversations
message_field_role: from
message_field_content: value
split: "train[:3%]"
```
## Run Training To use more VRAM (~14GB) and improve gradient signal, increase `micro_batch_size: 2`
```bash (adjust `gradient_accumulation_steps: 4` to keep effective batch at 8).
export PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True
axolotl train ~/human_chat_qlora.yml
```
Expected startup sequence:
1. Config validation + capability detection (shows `sm_120`)
2. Dataset tokenization (~65k samples, ~30 seconds)
3. `Loading weights: 100% 291/291`
4. `trainable params: 167,772,160 || all params: 8,198,033,408 || trainable%: 2.05`
5. Initial eval: loss ~0.81, perplexity ~2.25, VRAM ~8.5GB
6. Training steps at ~2.6 it/s, VRAM ~9-10GB
--- ---
## Common Pitfalls Encountered ## Common Pitfalls
| Problem | Cause | Fix | | Problem | Cause | Fix |
|---|---|---| |---|---|---|
@@ -167,12 +201,13 @@ Expected startup sequence:
| `CUDA_HOME not set` | CUDA toolkit not installed | `conda install cuda-toolkit` from nvidia channel | | `CUDA_HOME not set` | CUDA toolkit not installed | `conda install cuda-toolkit` from nvidia channel |
| `CUDA version mismatch 13.2 vs 12.8` | Conda nvcc is 13.2, torch was cu128 | Reinstall torch with `--index-url .../cu132` | | `CUDA version mismatch 13.2 vs 12.8` | Conda nvcc is 13.2, torch was cu128 | Reinstall torch with `--index-url .../cu132` |
| `torchaudio` not found for cu132 | No cu132 wheel exists | Skip torchaudio — not needed | | `torchaudio` not found for cu132 | No cu132 wheel exists | Skip torchaudio — not needed |
| `src refspec main does not match` | Fork default branch is `activeblue/main` | `git push origin activeblue/main` |
| flash-attn compile is slow | Single-threaded by default | Set `MAX_JOBS=<cpu_count>` before pip install | | flash-attn compile is slow | Single-threaded by default | Set `MAX_JOBS=<cpu_count>` before pip install |
| `nvcc fatal: Unsupported gpu architecture 'compute_50'` | bitsandbytes CMakeLists.txt hardcodes sm_50; CUDA 13.2 dropped it | Patch CMakeLists.txt (see step 7 above) | | `nvcc fatal: Unsupported gpu architecture 'compute_50'` | bitsandbytes CMakeLists.txt hardcodes sm_50; CUDA 13.2 dropped it | Patch CMakeLists.txt (see step 8 above) |
| `CUDA Capabilities Selected: 50;52;...` (ignores sm_120) | cmake >= 3.23 built-in arch list lacks sm_120 | Add `set(CMAKE_CUDA_ARCHITECTURES_ALL 120)` before foreach loop | | `CUDA Capabilities Selected: 50;52;...` (ignores -D flag) | cmake >= 3.23 built-in arch list lacks sm_120; CMakeLists.txt overrides -D | Insert `set(CMAKE_CUDA_ARCHITECTURES_ALL 120)` before foreach loop |
| `BackendUnavailable: scikit_build_core` | pip install of bnb tries to rebuild | Copy .so directly to site-packages instead | | `BackendUnavailable: scikit_build_core` | pip install of bnb triggers cmake rebuild | Copy .so directly to site-packages instead |
| `torch.OutOfMemoryError` during eval | logits tensor (batch x 4096 x 128k vocab) too large | Set `sequence_len: 2048`, `micro_batch_size: 1` | | `torch.OutOfMemoryError` during eval | logits tensor (batch x 4096 x 128k vocab) too large | Set `sequence_len: 2048`, `micro_batch_size: 1` |
| `type: sharegpt` deprecation warning | axolotl removed sharegpt type | Use `type: chat_template` with field mappings | | `type: sharegpt` deprecation warning | axolotl removed sharegpt type | Use `type: chat_template` with field mappings |
| `flash_attention: true` deprecation | Old config key removed | Use `attn_implementation: flash_attention_2` | | `flash_attention: true` deprecation | Old config key removed | Use `attn_implementation: flash_attention_2` |
| Capybara dataset `field_messages null` | Capybara uses input/output format, not conversations | Switch to SlimOrca or OpenHermes-2.5 | | Capybara dataset `field_messages null` | Capybara uses input/output format, not conversations | Switch to SlimOrca or OpenHermes-2.5 |
| Ollama loads model mid-training | Ollama is enabled and receives a request | `sudo systemctl stop ollama` before training |
| Training slower than expected (~3.5h not 19min) | The fast it/s on screen is the eval loop, not training | Normal — training includes backward pass and optimizer |