Hugging Face's logo

nraptisss
/
tmf921-intent-training

PEFT
qlora
sft
trl
qwen3
tmf921
intent-based-networking
network-slicing
rtx-6000-ada
ml-intern
Model card Files
xet
 Community
tmf921-intent-training / README.md
nraptisss's picture
nraptisss
Update ML Intern artifact metadata
0a2edf2 verified
preview code
|
Raw
History Blame
7.49 kB
metadata 
license: apache-2.0
tags:
  - qlora
  - sft
  - trl
  - peft
  - qwen3
  - tmf921
  - intent-based-networking
  - network-slicing
  - rtx-6000-ada
  - ml-intern
base_model:
  - Qwen/Qwen3-8B
datasets:
  - nraptisss/TMF921-intent-to-config-research-sota

TMF921 Intent-to-Config Training + Evaluation

Training and evaluation repo for nraptisss/TMF921-intent-to-config-research-sota on a single RTX 6000 Ada 48/50GB server.

The default recipe is Qwen3-8B + QLoRA NF4 + TRL SFTTrainer + PEFT LoRA.

Why this recipe

  • Dataset rows were audited with Qwen/Qwen3-8B chat-template tokenization.
  • Source max length: 1,316 tokens, p99: 1,300, so max_length=2048 is safe.
  • QLoRA NF4 + double quant follows the QLoRA recipe for fitting large models on one 48GB-class GPU.
  • LoRA uses target_modules="all-linear", recommended for QLoRA-style training.
  • assistant_only_loss=True trains only the JSON/config response tokens.
  • Evaluation is split by in-distribution and OOD splits; do not report only a single merged score.

Hardware target

Recommended server:

  • GPU: NVIDIA RTX 6000 Ada, 48GB/50GB VRAM
  • RAM: 64GB+
  • Disk: 200GB+ free
  • CUDA-compatible PyTorch

Default effective batch size:

per_device_train_batch_size = 2
gradient_accumulation_steps = 8
effective batch size = 16
max_length = 2048

If OOM occurs, preserve the effective batch size by changing:

per_device_train_batch_size: 1
gradient_accumulation_steps: 16

Do not reduce max_length unless you intentionally want a different training task.

Quick start with nohup, unique run dirs, and resumable checkpoints 

git clone https://huggingface.co/nraptisss/tmf921-intent-training
cd tmf921-intent-training

python -m venv .venv
source .venv/bin/activate
python -m pip install -U pip
bash scripts/install_rtx6000ada.sh
python scripts/check_gpu.py

export HF_TOKEN=hf_...
export CUDA_VISIBLE_DEVICES=0
export PYTHONPATH="$PWD/src"
export TOKENIZERS_PARALLELISM=false

# Optional Trackio dashboard
# export TRACKIO_SPACE_ID=nraptisss/tmf921-trackio

bash scripts/nohup_new_run.sh

The helper creates a fresh run directory every time:

runs/qwen3-8b-qlora-YYYYMMDD-HHMMSS/
  configs/config.yaml
  logs/train.log
  outputs/adapter/checkpoint-*/
  eval/

Monitor:

RUN_DIR=runs/qwen3-8b-qlora-YYYYMMDD-HHMMSS
bash scripts/status_run.sh "$RUN_DIR"
tail -f "$RUN_DIR/logs/train.log"
watch -n 2 nvidia-smi

Resume after crash/reboot:

cd tmf921-intent-training
source .venv/bin/activate
export HF_TOKEN=hf_...
export CUDA_VISIBLE_DEVICES=0
export PYTHONPATH="$PWD/src"

bash scripts/nohup_resume.sh runs/qwen3-8b-qlora-YYYYMMDD-HHMMSS

Evaluate after training:

bash scripts/nohup_eval.sh runs/qwen3-8b-qlora-YYYYMMDD-HHMMSS

Manual training command, if you do not want nohup:

python scripts/train_qlora.py \
  --config configs/rtx6000ada_qwen3_8b_qlora.yaml

Optional Trackio monitoring

The training script uses the native Transformers/TRL Trackio integration when project is set in the config.

Set a Trackio Space if desired:

export TRACKIO_SPACE_ID=nraptisss/tmf921-trackio

Or edit:

project: tmf921-intent-sft
trackio_space_id: nraptisss/tmf921-trackio

The trainer logs plain-text loss lines with:

disable_tqdm=True
logging_strategy="steps"
logging_first_step=True
report_to="trackio"

A callback emits Trackio alerts for NaN/Inf loss, high gradient norm, and high eval loss.

Configs

Recommended

configs/rtx6000ada_qwen3_8b_qlora.yaml

Key settings:

model_name_or_path: Qwen/Qwen3-8B
dataset_name: nraptisss/TMF921-intent-to-config-research-sota
train_split: train_sota
eval_split: validation
max_length: 2048
assistant_only_loss: true
load_in_4bit: true
lora_r: 64
lora_alpha: 16
lora_target_modules: all-linear
learning_rate: 0.0002
optim: paged_adamw_32bit
bf16: true
push_to_hub: true

Experimental 14B

configs/rtx6000ada_qwen3_14b_qlora_experimental.yaml

Use only after the 8B run, and expect tighter memory.

Evaluation

After training adapters:

python scripts/evaluate_model.py \
  --model Qwen/Qwen3-8B \
  --adapter outputs/qwen3-8b-tmf921-qlora \
  --dataset nraptisss/TMF921-intent-to-config-research-sota \
  --output_dir outputs/qwen3-8b-tmf921-qlora/eval \
  --load_in_4bit

Evaluated splits by default:

  • test_in_distribution
  • test_template_ood
  • test_use_case_ood
  • test_sector_ood
  • test_adversarial

Metrics:

  • JSON parse rate
  • canonical JSON exact match
  • field precision / recall / F1
  • slice/SST diagnostic pass
  • KPI text-presence diagnostic pass
  • adversarial status pass
  • stratified metrics by target_layer, slice_type, and lifecycle_operation

Outputs:

outputs/.../eval/all_metrics.json
outputs/.../eval/<split>/metrics.json
outputs/.../eval/<split>/predictions.json

Merge adapter for deployment 

python scripts/merge_adapter.py \
  --base_model Qwen/Qwen3-8B \
  --adapter outputs/qwen3-8b-tmf921-qlora \
  --output_dir outputs/qwen3-8b-tmf921-merged

Push merged model:

python scripts/merge_adapter.py \
  --base_model Qwen/Qwen3-8B \
  --adapter nraptisss/Qwen3-8B-TMF921-Intent-QLoRA-ResearchSOTA \
  --output_dir outputs/merged \
  --push_to_hub \
  --hub_model_id nraptisss/Qwen3-8B-TMF921-Intent-Merged

Scientific reporting protocol

For research papers/reports, report at least:

  1. validation loss,
  2. test_in_distribution metrics,
  3. test_template_ood metrics,
  4. test_use_case_ood metrics,
  5. test_sector_ood metrics,
  6. test_adversarial metrics,
  7. per-target-layer field F1,
  8. JSON parse rate,
  9. exact-match rate,
  10. rare-class metrics for lifecycle operations and adversarial categories.

Do not claim production standards compliance from JSON validity alone. Official TMF921/3GPP/ETSI/CAMARA/O-RAN validators are still needed for schema-level certification.

Files 

configs/
  rtx6000ada_qwen3_8b_qlora.yaml
  rtx6000ada_qwen3_14b_qlora_experimental.yaml
scripts/
  train_qlora.py
  evaluate_model.py
  merge_adapter.py
  run_rtx6000ada.sh
  nohup_new_run.sh
  nohup_resume.sh
  nohup_eval.sh
  status_run.sh
src/tmf921_train/
  utils.py
requirements.txt

References

Generated by ML Intern

This model repository was generated by ML Intern, an agent for machine learning research and development on the Hugging Face Hub.

Usage 

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = 'nraptisss/tmf921-intent-training'
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)

For non-causal architectures, replace AutoModelForCausalLM with the appropriate AutoModel class.