Lower vram usage for flux 2 text encoder.

comfy/sd1_clip.py

@@ -90,7 +90,6 @@ def __init__(self, device="cpu", max_length=77,
                  special_tokens={"start": 49406, "end": 49407, "pad": 49407}, layer_norm_hidden_state=True, enable_attention_masks=False, zero_out_masked=False,
                  return_projected_pooled=True, return_attention_masks=False, model_options={}):  # clip-vit-base-patch32
         super().__init__()
-        assert layer in self.LAYERS
 
         if textmodel_json_config is None:
             textmodel_json_config = os.path.join(os.path.dirname(os.path.realpath(__file__)), "sd1_clip_config.json")
@@ -164,7 +163,7 @@ def freeze(self):
     def set_clip_options(self, options):
         layer_idx = options.get("layer", self.layer_idx)
         self.return_projected_pooled = options.get("projected_pooled", self.return_projected_pooled)
-        if self.layer == "all":
+        if isinstance(self.layer, list) or self.layer == "all":
             pass
         elif layer_idx is None or abs(layer_idx) > self.num_layers:
             self.layer = "last"
@@ -266,7 +265,9 @@ def forward(self, tokens):
         if self.enable_attention_masks:
             attention_mask_model = attention_mask
 
-        if self.layer == "all":
+        if isinstance(self.layer, list):
+            intermediate_output = self.layer
+        elif self.layer == "all":
             intermediate_output = "all"
         else:
             intermediate_output = self.layer_idx

comfy/text_encoders/flux.py

@@ -138,7 +138,7 @@ def tokenize_with_weights(self, text, return_word_ids=False, llama_template=None
         return tokens
 
 class Mistral3_24BModel(sd1_clip.SDClipModel):
-    def __init__(self, device="cpu", layer="all", layer_idx=None, dtype=None, attention_mask=True, model_options={}):
+    def __init__(self, device="cpu", layer=[10, 20, 30], layer_idx=None, dtype=None, attention_mask=True, model_options={}):
         textmodel_json_config = {}
         num_layers = model_options.get("num_layers", None)
         if num_layers is not None:
@@ -154,7 +154,7 @@ def __init__(self, device="cpu", dtype=None, model_options={}, name="mistral3_24
     def encode_token_weights(self, token_weight_pairs):
         out, pooled, extra = super().encode_token_weights(token_weight_pairs)
 
-        out = torch.stack((out[:, 10], out[:, 20], out[:, 30]), dim=1)
+        out = torch.stack((out[:, 0], out[:, 1], out[:, 2]), dim=1)
         out = out.movedim(1, 2)
         out = out.reshape(out.shape[0], out.shape[1], -1)
         return out, pooled, extra

comfy/text_encoders/llama.py

@@ -434,16 +434,21 @@ def forward(self, x, attention_mask=None, embeds=None, num_tokens=None, intermed
 
         intermediate = None
         all_intermediate = None
+        only_layers = None
         if intermediate_output is not None:
-            if intermediate_output == "all":
+            if isinstance(intermediate_output, list):
+                all_intermediate = []
+                only_layers = set(intermediate_output)
+            elif intermediate_output == "all":
                 all_intermediate = []
                 intermediate_output = None
             elif intermediate_output < 0:
                 intermediate_output = len(self.layers) + intermediate_output
 
         for i, layer in enumerate(self.layers):
             if all_intermediate is not None:
-                all_intermediate.append(x.unsqueeze(1).clone())
+                if only_layers is None or (i in only_layers):
+                    all_intermediate.append(x.unsqueeze(1).clone())
             x = layer(
                 x=x,
                 attention_mask=mask,
@@ -457,7 +462,8 @@ def forward(self, x, attention_mask=None, embeds=None, num_tokens=None, intermed
             x = self.norm(x)
 
         if all_intermediate is not None:
-            all_intermediate.append(x.unsqueeze(1).clone())
+            if only_layers is None or ((i + 1) in only_layers):
+                all_intermediate.append(x.unsqueeze(1).clone())
 
         if all_intermediate is not None:
             intermediate = torch.cat(all_intermediate, dim=1)