LLM: enable hybrid-mode export via num_sharding=1 — TTFT 2.9s → 113ms

Re-exported Qwen3-4B in hybrid mode (prefill_forward + kv_forward) with
num_sharding=1 after discovering that sharding=2 produces a multi-context
.pte that the LlmModule loader cannot restore (error 5010 "Context group 1
does not exist"). Single-context hybrid .pte loads cleanly through the JNI
runner and the auto-detected eval_mode=1 path.

The peak RAM during export hit 49 GB, which is why sharding=2 was used
originally — the /swapfile (192 GB) now absorbs it. Compile wall time with
sharding=1 + hybrid is ~73 min (two graphs) vs ~30 min for sharding=2 +
kv-only (one graph).

End-to-end on tablet, same 'Bonjour, comment vas-tu ?' prompt:
  Before (kv-only, short prompt):  TTFT 2865 ms, total 4034 ms
  After  (hybrid, short prompt):   TTFT  113 ms, total 1471 ms
  Gain: -2752 ms TTFT (96% reduction, 25× faster)
  Response: "Bonjour ! Je vais bien, merci de me demander. Comment vas-tu ?"

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

executorch-patches/qwen3_4b_decoder.patch

@@ -1,5 +1,5 @@
 diff --git a/examples/qualcomm/oss_scripts/llama/__init__.py b/examples/qualcomm/oss_scripts/llama/__init__.py
-index 963db6e..953dc4c 100644
+index 963db6e..9ccfdd0 100644
 --- a/examples/qualcomm/oss_scripts/llama/__init__.py
 +++ b/examples/qualcomm/oss_scripts/llama/__init__.py
 @@ -25,9 +25,14 @@ from executorch.examples.models.granite import (
@@ -20,7 +20,7 @@ index 963db6e..953dc4c 100644
  from executorch.examples.models.qwen2_5 import (
      convert_weights as convert_qwen2_5_weights,
  )
-@@ -479,6 +484,34 @@ class Qwen3_1_7B(LLMModelConfig):
+@@ -479,6 +484,37 @@ class Qwen3_1_7B(LLMModelConfig):
      quant_recipe = Qwen3_1_7BQuantRecipe
  
  
@@ -40,10 +40,13 @@ index 963db6e..953dc4c 100644
 +    convert_weights = convert_qwen3_weights
 +    transform_weight = False
 +    instruct_model = True
-+    # Bumped to 2 to halve peak host RAM during QNN compile (4B at sharding=1
-+    # OOMed on a 62 GB box, peak anon-rss 46 GB). At sharding=2 each shard
-+    # compile fits comfortably; runner stitches them at load time.
-+    num_sharding = 2
++    # num_sharding=1 for hybrid mode: sharding=2 produces a multi-context
++    # .pte (2 graphs × 2 shards = 4 contexts) that the LlmModule load path
++    # can't restore (error 5010 "Context group 1 does not exist"). With
++    # sharding=1 the hybrid export needs ~46 GB RAM peak — the 192 GB swap
++    # on /swapfile handles this; compile takes ~80 min wall but completes
++    # cleanly. Single-context .pte loads fine through the JNI runner.
++    num_sharding = 1
 +    masked_softmax = True
 +    seq_mse_candidates = 0
 +    r1 = False