don't train if eval split is too small

src/axolotl/core/trainer_builder.py

@@ -658,7 +658,9 @@ def build(self, total_num_steps):
             self.cfg.sample_packing if self.cfg.sample_packing else False
         )
         training_arguments_kwargs["eval_sample_packing"] = (
-            self.cfg.sample_packing if self.cfg.sample_packing else False
+            self.cfg.sample_packing
+            if self.cfg.eval_sample_packing is not False
+            else False
         )
         training_arguments_kwargs[
             "sample_packing_seq_len_multiplier"

src/axolotl/utils/data.py

@@ -79,6 +79,14 @@ def prepare_dataset(cfg, tokenizer):
         train_dataset, eval_dataset = process_datasets_for_packing(
             cfg, train_dataset, eval_dataset, tokenizer
         )
+
+    if eval_dataset and cfg.sample_packing and cfg.eval_sample_packing is not False:
+        total_eval_steps = calculate_total_num_steps(cfg, eval_dataset, update=False)
+        if total_eval_steps == 0:
+            raise ValueError(
+                "eval dataset split is too small for sample_packing. You should set `eval_sample_packing: False`. "
+            )
+
     if cfg.max_steps:
         total_num_steps = min(
             calculate_total_num_steps(cfg, train_dataset), cfg.max_steps

src/axolotl/utils/samplers/multipack.py

@@ -182,7 +182,7 @@ def _len_est(self):
 
         # shave off 1% + 1 for dealing with variance in packing from random sampler to sampler
         return max(
-            1,
+            0,
             (
                 world_size
                 * math.floor(

src/axolotl/utils/trainer.py

@@ -141,7 +141,7 @@ def process_datasets_for_packing(cfg, train_dataset, eval_dataset, tokenizer):
     return train_dataset, eval_dataset
 
 
-def calculate_total_num_steps(cfg, train_dataset):
+def calculate_total_num_steps(cfg, train_dataset, update=True):
     if not cfg.total_num_tokens:
         total_num_tokens = np.sum(
             train_dataset.data.column("input_ids")
@@ -150,7 +150,8 @@ def calculate_total_num_steps(cfg, train_dataset):
             .values
         )
         LOG.debug(f"total_num_tokens: {total_num_tokens}", main_process_only=True)
-        cfg.total_num_tokens = total_num_tokens
+        if update:
+            cfg.total_num_tokens = total_num_tokens
 
     if not cfg.total_supervised_tokens:
         total_supervised_tokens = (
@@ -163,7 +164,8 @@ def calculate_total_num_steps(cfg, train_dataset):
             f"`total_supervised_tokens: {total_supervised_tokens}`",
             main_process_only=True,
         )
-        cfg.total_supervised_tokens = total_supervised_tokens
+        if update:
+            cfg.total_supervised_tokens = total_supervised_tokens
 
     if cfg.sample_packing:
         # we have to drop anything longer then sequence len otherwise
@@ -232,7 +234,8 @@ def calc_sample_packing_eff_est(estimates: List[float]):
             sample_packing_eff_est = (
                 math.ceil(sample_packing_actual_eff_all * 100.0) / 100.0
             )
-            cfg.sample_packing_eff_est = sample_packing_eff_est
+            if update:
+                cfg.sample_packing_eff_est = sample_packing_eff_est
             LOG.debug(
                 f"sample_packing_eff_est: {cfg.sample_packing_eff_est}",
                 main_process_only=True,