qwen4exp: fix seq_cp, block position keying, mtmd input, cuda abort, add tests (#27941)

* qwen4exp: follow up fixes

* -kvu NaN collapse fix

Assisted-by: Claude

* indexer cache ext.x/ext.y restore fix

Assisted-by: Claude

* kv-cells: rename seq_set to seq_get_all

seq_get is already taken by the single-id getter, so the suggested name
cannot be overloaded on return type alone.

Assisted-by: Claude

* memory-hybrid-idx: implement set_input_qsa on the memory class

The context held the whole implementation, where the pattern elsewhere is a
thin context forwarding to the memory class, as llama_kv_cache_context does
for set_input_kq_mask. The body reads no context state, so it moves unchanged
and the context keeps a forwarder.

Also shortens the seq_get_all comment as suggested.

* tests: check that a sequence state survives a save/restore round-trip

Saves seq 0, erases it, restores the blob and saves again, requiring the two
blobs to match. Compares blobs rather than generated text, which cannot see a
field dropped on the way back in.

Note this passes on master for qwen4exp, so it does not demonstrate the
ext.x/ext.y drop this PR fixes; reaching that needs 2D mrope content.

* tests: give the synthetic qwen4exp a PLE so the state test bites

has_cell_ext() is n_pos_per_embd() > 1 || ple_n_heads > 0, and the indexer
cache sets rope_type = NONE, so without a PLE it serializes no cell ext at
all and the round-trip test cannot see a dropped ext.x/ext.y. With one,
removing the ext_set restore in state_read_meta fails the test: 198 of
335692 bytes differ, first at offset 282092.

Loading such a model needed two fixes:

- the row count of per_layer_token_embd came from require_weight(), which a
  model synthesised from metadata alone has no file to answer. Derive it
  from the head ranges and prefer the file's padded count where there is one.
- the PLE conv history is a row of the recurrent cache, so a PLE on a full
  attention layer dereferenced a null p_l. Reject it at load time instead.

The meta mirror is skipped for qwen4exp. It returned NaN logits before this
fixture carried a PLE, which the nmse check passes since a NaN comparison is
false, and aborts with one. -sm tensor on real devices works.

Assisted-by: Claude

* llama: disable -sm tensor for qwen4exp

test-llama-archs skipped the tensor split for this arch from inside the
test, so the arch still advertised support it does not have. Declare it in
llm_arch_supports_sm_tensor instead and drop the test-side exception; the
existing llm_arch_supports_sm_tensor branch then does the skipping.

Assisted-by: Claude

src/llama-memory-hybrid-idx.h

@@ -75,6 +75,18 @@ public:
 
     llama_kv_cache * get_mem_idx() const;   // nullptr when the model carries no indexer
 
+    // block-compressed sparse attention (qwen4exp QSA) over the cells of the indexer cache.
+    // Blocks cut the position line, not the cell array, so no caller assumes a contiguous layout:
+    //   cell_blk  I32 [n_kv, ns]           block each cell belongs to
+    //   blk_cells I32 [ratio*n_blocks, ns] cells making up each block
+    //   blk_pos   I32 [4*n_blocks*ns]      mrope position rows of each block's first token
+    //   bias      F32 [n_kv, n_tokens/ns, ns] -inf where invisible, large where always visible
+    // blk_bias asks for the bias per block instead: [n_blocks, n_tokens/ns, ns]
+    // the caller then adds the attention mask, the only part of the bias that varies within a block
+    void set_input_qsa(ggml_tensor * cell_blk, ggml_tensor * blk_cells, ggml_tensor * blk_pos,
+                       ggml_tensor * bias, const llama_ubatch * ubatch, uint32_t ratio,
+                       bool blk_bias) const;
+
 private:
     // forget seq_id (all of it if seq_id < 0) in every cache at once, so a failed restore cannot leave the caches out of step
     // seq_id < 0 drops the whole context, as the caches themselves do on a failed restore
@@ -123,20 +135,12 @@ public:
     // llama_memory_hybrid_idx_context specific API
     //
 
-    // nullptr with no indexer, and for the update context, which builds no sparse graph
+    // nullptr with no indexer
     const llama_kv_cache_context * get_idx() const;
 
     // streams in the current slot info, the `ns` of get_k/get_v; 1 if unified
     uint32_t get_n_stream() const;
 
-    // block-compressed sparse attention (qwen4exp QSA) over the cells of the indexer cache.
-    // Blocks cut the position line, not the cell array, so no caller assumes a contiguous layout:
-    //   cell_blk  I32 [n_kv, ns]           block each cell belongs to
-    //   blk_cells I32 [ratio*n_blocks, ns] cells making up each block
-    //   blk_pos   I32 [4*n_blocks*ns]      mrope position rows of each block's first token
-    //   bias      F32 [n_kv, n_tokens/ns, ns] -inf where invisible, large where always visible
-    // blk_bias asks for the bias per block instead: [n_blocks, n_tokens/ns, ns]
-    // the caller then adds the attention mask, the only part of the bias that varies within a block
     void set_input_qsa(ggml_tensor * cell_blk, ggml_tensor * blk_cells, ggml_tensor * blk_pos,
                        ggml_tensor * bias, const llama_ubatch * ubatch, uint32_t ratio,
                        bool blk_bias) const;
@@ -148,7 +152,7 @@ private:
     // declared first, so it is initialised while sinfos_idx is still intact
     const std::vector<uint32_t> ns_ubatch;
 
-    // null unless the model has an indexer and this is a batch or full context
+    // null unless the model has an indexer
     const llama_memory_context_ptr ctx_idx;
 
     // mirrors the base class's ubatch cursor, which is private there