refactor(qwen35): split scheduler backend into scheduler/backend.rs

pegainfer-qwen35/Cargo.toml

@@ -19,6 +19,7 @@ rand = { workspace = true }
 safetensors = { workspace = true }
 serde = { workspace = true }
 serde_json = { workspace = true }
+thiserror = { workspace = true }
 tokio = { workspace = true, features = ["sync"] }
 
 [dev-dependencies]

pegainfer-qwen35/src/batch_decode.rs

@@ -102,9 +102,9 @@ impl Qwen35Model {
         bufs: &mut BatchDecodeBuffers35,
     ) -> Result<()> {
         let eps = self.config.rms_norm_eps;
-        let tp = self.tensor_parallel;
-        let num_attention_heads = self.config.local_num_attention_heads(tp);
-        let num_key_value_heads = self.config.local_num_key_value_heads(tp);
+        let geom = self.geometry;
+        let num_attention_heads = geom.local_num_attention_heads();
+        let num_key_value_heads = geom.local_num_key_value_heads();
 
         ops::gemm_into(&self.ctx, &attn.q_proj, &bufs.normed, &mut bufs.q_full);
         ops::gemm_into(&self.ctx, &attn.k_proj, &bufs.normed, &mut bufs.k_attn);

pegainfer-qwen35/src/batch_decode_graph.rs

@@ -6,7 +6,7 @@ use pegainfer_core::kv_pool::KvPool;
 use pegainfer_core::tensor::DeviceContext;
 
 use super::config::Config35;
-use super::config::TensorParallelConfig;
+use super::config::LocalGeometry;
 use super::decode_buffers::BatchDecodeBuffers35;
 use super::recurrent_state::LinearStatePointerTables;
 use super::recurrent_state::RecurrentState;
@@ -61,7 +61,7 @@ impl BatchDecodeGraphState {
     pub(crate) fn with_capacity(
         ctx: &DeviceContext,
         config: &Config35,
-        tensor_parallel: TensorParallelConfig,
+        geometry: LocalGeometry,
         kv_pool: &KvPool,
         max_batch: usize,
     ) -> Result<Self> {
@@ -71,7 +71,7 @@ impl BatchDecodeGraphState {
         let buffers = BatchDecodeBuffers35::new(
             ctx,
             config,
-            tensor_parallel,
+            geometry,
             max_batch,
             max_total_pages,
             padding_page_id,