better rescaling for temperatures

src/axolotl/core/trainer_builder.py

@@ -680,6 +680,8 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
 
         if self.cfg.kd_ce_alpha is not None:
             training_arguments_kwargs["kd_ce_alpha"] = self.cfg.kd_ce_alpha
+        if self.cfg.kd_alpha is not None:
+            training_arguments_kwargs["kd_alpha"] = self.cfg.kd_alpha
 
         training_args_cls = (
             AxolotlTrainingArguments

src/axolotl/core/trainers/kd.py

@@ -16,6 +16,7 @@ def kd_loss_function(
     target_logprobs,
     target_mask,
     num_items_in_batch: Optional[int] = None,
+    kd_temperature: float = 1.0,
 ):
     # teacher_mask: [B, teacher_seq_len, K], where 1 indicates a valid token and 0 indicates padding
 
@@ -28,9 +29,15 @@ def kd_loss_function(
     ]  # [B, teacher_seq_len, vocab_size]
 
     # Gather student logits for teacher's top-K tokens
+    #  shape -> [B, teacher_seq_len, K]
     student_logits_topk = torch.gather(
         student_logits_for_kd, dim=-1, index=target_token_ids
-    )  # [B, teacher_seq_len, K]
+    )
+
+    # Apply KD temperature to student’s logits:
+    #  z_s(T) = z_s / T
+    if kd_temperature != 1.0:
+        student_logits_topk = student_logits_topk / kd_temperature
 
     # Convert student top-k logits to logprobs
     student_logprobs_topk = student_logits_topk - torch.logsumexp(
@@ -53,6 +60,10 @@ def kd_loss_function(
     kd_loss_per_token = teacher_probs * (target_logprobs - student_logprobs_topk)
     kd_loss = kd_loss_per_token.sum()
 
+    # 9) Multiply by T^2 (classical KD scaling)
+    if kd_temperature != 1.0:
+        kd_loss = kd_loss * (kd_temperature**2)
+
     # Normalize by number of items or mean over valid tokens
     if num_items_in_batch is not None:
         # If you know how many items should be considered in the batch
@@ -129,7 +140,8 @@ class AxolotlKDTrainer(AxolotlTrainer):
 
         # optionally combine with CE loss
         if self.args.kd_ce_alpha > 0:
-            loss = self.args.kd_ce_alpha * outputs["loss"] + loss_kd
+            kd_alpha = self.args.kd_alpha
+            loss = self.args.kd_ce_alpha * outputs["loss"] + kd_alpha * loss_kd
         else:
             loss = loss_kd
 
@@ -166,6 +178,7 @@ class AxolotlKDTrainer(AxolotlTrainer):
             target_logprobs,
             target_mask,
             num_items_in_batch=num_items_in_batch,
+            kd_temperature=self.args.kd_temperature,
         )
 
         if self.args.kd_ce_alpha > 0:

src/axolotl/core/training_args.py

@@ -172,7 +172,19 @@ class AxolotlTrainingMixins:
     kd_ce_alpha: Optional[float] = field(
         default=None,
         metadata={
-            "help": "The alpha parameter for SFT cross entropy loss when using KD"
+            "help": "The alpha scaling parameter for SFT cross entropy loss when using KD"
+        },
+    )
+
+    kd_alpha: Optional[float] = field(
+        default=1.0,
+        metadata={"help": "The alpha scaling parameter for KD loss"},
+    )
+
+    kd_temperature: Optional[float] = field(
+        default=1.0,
+        metadata={
+            "help": "the temperature parameter for KL divergence loss when using KD"
         },
     )
 

src/axolotl/prompt_strategies/chat_template.py

@@ -474,10 +474,12 @@ class ChatTemplateStrategyWithKD(ChatTemplateStrategy):
         roles_to_train=None,
         train_on_eos=None,
         logprobs_field="logprobs",
-        temperature=1.0,
+        gen_temperature=1.0,
+        kd_temperature=1.0,
     ):
         self.logprobs_field = logprobs_field
-        self.temperature = temperature
+        self.gen_temperature = gen_temperature
+        self.kd_temperature = kd_temperature
 
         super().__init__(
             prompter,
@@ -531,14 +533,25 @@ class ChatTemplateStrategyWithKD(ChatTemplateStrategy):
                 position_logprobs, dtype=torch.float
             )
 
-            # Apply temperature scaling at data load time
+            if self.kd_temperature != self.gen_temperature:
-            # log p_k^(T) = (log p_k / T) - logsumexp(log p_j / T)
+                #
-            position_logprobs_tensor = position_logprobs_tensor / self.temperature
+                # Now we have distribution at T1 in log form, i.e. log p_{T1}(k).
-            # normalize to probabilities so they sum up to 1
+                # Next, re-scale to T2 = self.kd_temperature via exponent-based trick
-            position_logprobs_tensor = position_logprobs_tensor - torch.logsumexp(
+                # p_{T2}(k) = [p_{T1}(k)]^(T1 / T2) / Z
-                position_logprobs_tensor, dim=0, keepdim=True
+                #
-            )
+                # Convert from log to probability
+                teacher_probs_t1 = position_logprobs_tensor.exp()
+                # Exponentiate by factor (T1 / T2)
+                exponent = self.gen_temperature / self.kd_temperature
+                teacher_probs_t2 = teacher_probs_t1**exponent
+                # Re-normalize
+                teacher_probs_t2 = teacher_probs_t2 / teacher_probs_t2.sum(
+                    dim=0, keepdim=True
+                )
+                # Convert back to log
+                position_logprobs_tensor = torch.log(teacher_probs_t2)
 
+            # Now we have log p_{teacher, T2}(k) stored in position_logprobs_tensor
             position_logprobs_scaled = position_logprobs_tensor.tolist()
 
             target_logprobs.append(position_logprobs_scaled)
@@ -593,13 +606,15 @@ def load(tokenizer, cfg, ds_cfg: Optional[Dict[str, Any]] = None, processor=None
     }
 
     strategy_cls = ChatTemplateStrategy
-    if logprobs_field := ds_cfg.get("logprobs_field"):
-        strategy_params["logprobs_field"] = logprobs_field
-    if temperature := ds_cfg.get("temperature"):
-        strategy_params["temperature"] = temperature
 
-    if cfg.trainer == "kd" or logprobs_field:
+    if cfg.trainer == "kd":
         strategy_cls = ChatTemplateStrategyWithKD
+        if logprobs_field := ds_cfg.get("logprobs_field"):
+            strategy_params["logprobs_field"] = logprobs_field
+        if gen_temperature := ds_cfg.get("temperature"):
+            strategy_params["gen_temperature"] = gen_temperature
+        if kd_temperature := cfg.get("kd_temperature"):
+            strategy_params["kd_temperature"] = kd_temperature
 
     strategy = strategy_cls(
         ChatTemplatePrompter(**prompter_params), tokenizer=tokenizer, **strategy_params

src/axolotl/utils/config/models/input/v0_4_1/__init__.py

@@ -178,6 +178,7 @@ class SFTDataset(BaseModel):
     message_field_training: Optional[str] = None
     message_field_training_detail: Optional[str] = None
     logprobs_field: Optional[str] = None
+    temperature: Optional[float] = None
     roles_to_train: Optional[List[str]] = None
     train_on_eos: Optional[str] = None
     roles: Optional[Dict[str, List[str]]] = None
@@ -626,6 +627,8 @@ class AxolotlInputConfig(
     kd_ce_alpha: Optional[
         float
     ] = None  # loss coefficient for cross-entropy loss during KD
+    kd_alpha: Optional[float] = None  # loss coefficient for KD loss
+    kd_temperature: Optional[float] = None  # temperature for sampling during KD
 
     datasets: Optional[conlist(Union[SFTDataset, DPODataset, KTODataset], min_length=1)] = None  # type: ignore
     test_datasets: Optional[conlist(Union[SFTDataset, DPODataset, KTODataset], min_length=1)] = None  # type: ignore