Rename core MiniMindBlock to Block for modality-agnostic naming

The transformer block is a generic primitive reused by lm/vlm/vam
backbones, so rename MiniMindBlock -> Block in core/block.py and update
all references (core/__init__, models/__init__, models/lm/model.py,
models/vam/model.py, lora comment).

Co-Authored-By: opencode <noreply@opencode.ai>

src/omni/core/__init__.py

@@ -2,7 +2,7 @@ from omni.core.norm import RMSNorm
 from omni.core.rope import precompute_freqs_cis, apply_rotary_pos_emb, repeat_kv
 from omni.core.attention import Attention
 from omni.core.mlp import FeedForward, MOEFeedForward
-from omni.core.block import MiniMindBlock
+from omni.core.block import Block
 
 __all__ = [
     "RMSNorm",
@@ -12,5 +12,5 @@ __all__ = [
     "Attention",
     "FeedForward",
     "MOEFeedForward",
-    "MiniMindBlock",
+    "Block",
 ]

src/omni/core/block.py

@@ -5,7 +5,7 @@ from omni.core.attention import Attention
 from omni.core.mlp import FeedForward, MOEFeedForward
 
 
-class MiniMindBlock(nn.Module):
+class Block(nn.Module):
     def __init__(self, layer_id: int, config: "MiniMindConfig"):
         super().__init__()
         self.self_attn = Attention(config)

src/omni/models/__init__.py

@@ -16,7 +16,7 @@ from omni.core import (
     Attention,
     FeedForward,
     MOEFeedForward,
-    MiniMindBlock,
+    Block,
     precompute_freqs_cis,
     apply_rotary_pos_emb,
 )
@@ -38,8 +38,7 @@ __all__ = [
     "Attention",
     "FeedForward",
     "MOEFeedForward",
-    "MiniMindBlock",
-    "MiniMindModel",
+    "Block",
     "precompute_freqs_cis",
     "apply_rotary_pos_emb",
 ]

src/omni/models/lm/lora.py

@@ -3,7 +3,7 @@ from torch import optim, nn
 
 # LoRA for MiniMind-series backbones.
 # Applies to every nn.Linear whose in_features == out_features, i.e. the
-# q/k/v/o_proj and gate/up/down_proj inside MiniMindBlock's Attention/MLP.
+# q/k/v/o_proj and gate/up/down_proj inside Block's Attention/MLP.
 # Because VLM/Omni reuse the same MiniMind Attention/MLP for their LLM
 # (thinker)主干, apply_lora also works on MiniMindVLM / MiniMindOmni, but only
 # touches the shared LLM layers -- vision/audio projectors and the speech

src/omni/models/lm/model.py

@@ -4,7 +4,7 @@ from torch import nn
 from transformers import PreTrainedModel, GenerationMixin
 from transformers.modeling_outputs import MoeCausalLMOutputWithPast
 
-from omni.core import RMSNorm, precompute_freqs_cis, MiniMindBlock, MOEFeedForward
+from omni.core import RMSNorm, precompute_freqs_cis, Block, MOEFeedForward
 from omni.models.lm.config import MiniMindConfig
 
 
@@ -15,7 +15,7 @@ class MiniMindModel(nn.Module):
         self.vocab_size, self.num_hidden_layers = config.vocab_size, config.num_hidden_layers
         self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size)
         self.dropout = nn.Dropout(config.dropout)
-        self.layers = nn.ModuleList([MiniMindBlock(l, config) for l in range(self.num_hidden_layers)])
+        self.layers = nn.ModuleList([Block(l, config) for l in range(self.num_hidden_layers)])
         self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
         freqs_cos, freqs_sin = precompute_freqs_cis(dim=config.head_dim, end=config.max_position_embeddings, rope_base=config.rope_theta, rope_scaling=config.rope_scaling)
         self.register_buffer("freqs_cos", freqs_cos, persistent=False)

src/omni/models/vam/model.py

@@ -10,7 +10,7 @@ from torch.nn import functional as F
 from transformers.modeling_outputs import MoeCausalLMOutputWithPast
 from transformers import SiglipVisionModel, SiglipImageProcessor, logging as hf_logging
 
-from omni.core import RMSNorm, precompute_freqs_cis, MiniMindBlock, MOEFeedForward
+from omni.core import RMSNorm, precompute_freqs_cis, Block, MOEFeedForward
 from omni.models.lm.config import MiniMindConfig
 from omni.models.lm.model import MiniMindForCausalLM
 from omni.models.vam.config import OmniConfig
@@ -53,7 +53,7 @@ class TalkerModule(nn.Module):
     def __init__(self, config: OmniConfig):
         super().__init__()
         self.talker_config = MiniMindConfig(hidden_size=config.talker_hidden_size, use_moe=config.use_moe)
-        self.layers = nn.ModuleList([MiniMindBlock(l, self.talker_config) for l in range(config.num_talker_hidden_layers)])
+        self.layers = nn.ModuleList([Block(l, self.talker_config) for l in range(config.num_talker_hidden_layers)])
         self.norm = RMSNorm(config.talker_hidden_size, eps=config.rms_norm_eps)
         self.lm_head = TalkerHead(config.talker_hidden_size, config.audio_vocab_size)
         self.embed_tokens = TalkerEmbedding(config.audio_vocab_size, config.talker_hidden_size)