Only Tensors of floating point dtype can require gradients int8 LoRA

A LoRA adapter was attached to a quantized (int8/int4) base layer and the training path tried to make the quantized weight itself require gradients. Only the floating-point adapter weights are trainable; the quantized base must stay frozen. This entry explains how to confirm the cause, apply the fix, and separate it from adjacent huggingface failures.

The literal signature is "Only Tensors of floating point dtype can require gradients int8 LoRA". It is a environment failure associated with Hugging Face Transformers, Accelerate, and PEFT. The line identifies the failing operation or subsystem, while the surrounding evidence decides whether it is the initiating fault or a downstream symptom.

A LoRA adapter was attached to a quantized (int8/int4) base layer and the training path tried to make the quantized weight itself require gradients. Only the floating-point adapter weights are trainable; the quantized base must stay frozen. The failure becomes visible at this call site because the operation first requires the missing resource, valid state, healthy peer, or correct result. Earlier log lines and a known-good control carry more causal value than the final wrapper exception.

• The workload stops or loses forward progress after emitting "Only Tensors of floating point dtype can require gradients int8 LoRA".
• A retry on the same configuration reproduces the failure because the causal state has not changed.
• The outer framework exception can hide the rank, node, allocation, or dependency that failed first.
• Increasing timeouts or reducing workload size can suppress the symptom without correcting the cause.

• Hugging Face Transformers, Accelerate, and PEFT
• production-shaped multi-accelerator workloads
• containerized and bare-metal deployments of the same stack

Apply the following checklist to a small reproduction: each box below is a positive signal that you are looking at this exact failure rather than a sibling in the same taxonomy.

• ✓Find the first occurrence of "Only Tensors of floating point dtype can require gradients int8 LoRA" and preserve at least 100 lines before it.
• ✓Identify which rank, node, device, or process emitted the first related warning.
• ✓print the trainable set, [n for n,p in model.named_parameters() if p.requires_grad]. Only lora_A/lora_B (and optionally norms and the head) should appear. Any base weight in that list is the bug.
• ✓Repeat the same input after the targeted change and require the signature to disappear.
• ✓Resume from job start only after the control passes.

Only Tensors of floating point dtype can require gradients int8 LoRA

Timestamps and exact values vary across runs, but the pattern. An info-level start, an early WARN, an ERROR carrying the symptom. Is the actual fingerprint you should alert on. The Denpex platform flags this combination automatically.

call prepare_model_for_kbit_training(model) after loading the quantized base and BEFORE get_peft_model(). It freezes the quantized weights and casts the layers that must stay in float. This changes the condition in the causal diagnosis instead of hiding the outer exception. The repeated control proves ownership before recovery from job start.

# Preserve evidence before restarting
rg -n -i 'error|exception|timeout|failed' <log-file>
nvidia-smi
python -m torch.utils.collect_env

# Find the exact signature in the complete log
rg -n -F -- "Only Tensors of floating point dtype can require gradients int8 LoRA" <log-file>

Adapt the snippet to your framework. The same pattern holds for PyTorch Lightning, Hugging Face Trainer, DeepSpeed, Megatron-LM, and vLLM training wrappers. Where the wrapper exposes a config flag (for examplelr_scheduler_type in Trainer), prefer the flag over the imperative API to keep the schedule declarative and reproducible.

Select the traceback or open the failed terminal, then run Denpex locally to see the initiating rank, collateral failures, exact fix, and verification command without uploading the log.

• A LoRA adapter was attached to a quantized (int8/int4) base layer and the training path tried to make the quantized weight itself require gradients. Only the floating-point adapter weights are trainable; the quantized base must stay frozen.
• The decisive evidence is the first log line that precedes "Only Tensors of floating point dtype can require gradients int8 LoRA" and differs from a healthy run.
• A nearby failure remains a competing hypothesis until a control separates configuration, capacity, transport, and hardware causes.