docs(recipes): document the offloaded layerwise export recipe

The recipe shipped undocumented; the parent branch's docs test now requires a
row and the count kept in step.

Signed-off-by: Fridah-nv <201670829+Fridah-nv@users.noreply.github.com>

Author: Fridah-nv

modelopt_recipes/ptq.md

@@ -29,7 +29,7 @@ supported combinations.
 ### The shipped recipes
 
 <details>
-<summary>All 25 <code>general/ptq/</code> recipes (click to expand)</summary>
+<summary>All 26 <code>general/ptq/</code> recipes (click to expand)</summary>
 
 | Recipe | Model body | KV cache | Calibration |
 |--------|-----------|----------|-------------|
@@ -49,6 +49,7 @@ supported combinations.
 | `nvfp4_experts_only-kv_fp8_layerwise` | NVFP4 W4A4, MoE experts only | FP8 (calibrated) | max, layerwise |
 | `nvfp4_experts_only-kv_fp8_layerwise_offload` | NVFP4 W4A4, MoE experts only | FP8 (calibrated) | max, layerwise (non-mutating, for disk offload) |
 | `nvfp4_experts_only-kv_fp8_layerwise_export` | NVFP4 W4A4, MoE experts only | FP8 (calibrated) | max, layerwise (exports each layer as it is calibrated) |
+| `nvfp4_experts_only-kv_fp8_layerwise_export_offload` | NVFP4 W4A4, MoE experts only | FP8 (calibrated) | max, layerwise, tuned for accelerate-offloaded models |
 | `nvfp4_experts_only_mse-kv_fp8_cast` | NVFP4 W4A4, MoE experts only | FP8 (constant amax) | MSE + FP8 sweep |
 | `nvfp4_experts_only_input_scale1-kv_fp8_cast` | NVFP4 W4A4, MoE experts only, expert `input_scale` pinned to 1.0 | FP8 (constant amax) | max (weights); expert activations uncalibrated |
 | `nvfp4_omlp_only-kv_fp8` | NVFP4 W4A4, o_proj + MLP/MoE | FP8 (calibrated) | max |