Qualify spec recommendation: eagle3 for short-context, ngram for long-document summarize (measured)

README.md

@@ -148,10 +148,12 @@ vllm serve <this-repo> --trust-remote-code \
 
 Since 2026-07-23 the recommended speculation is my
 [EAGLE-3 draft head](https://huggingface.co/NullSense/Nanbeige4.2-3B-EAGLE3):
-221 freeform / 244 summarize / 262 code tok/s in one config (TRITON_ATTN required
-with it; FlashInfer + spec drops CUDA graphs to piecewise on SM120). The previous
-ngram config (`{"method":"ngram","num_speculative_tokens":8,"prompt_lookup_max":4,
-"prompt_lookup_min":2}`) remains the head-less alternative for copy-heavy work.
+221 freeform / 244 short-passage summarize / 262 code tok/s in one config
+(TRITON_ATTN required with it; FlashInfer + spec drops CUDA graphs to piecewise on
+SM120). One measured exception: LONG-document summarization (multi-k-char inputs)
+still favors the ngram config (`{"method":"ngram","num_speculative_tokens":8,
+"prompt_lookup_max":4,"prompt_lookup_min":2}`): 231 vs 160 output tok/s on 16 real
+articles; verbatim copy-spans reward lookup drafting. Pick per dominant workload.
 
 - Thinking ON by default; pass `chat_template_kwargs: {"enable_thinking": false}`
   (+ `preserve_thinking: false` for chat) to disable. Sampling defaults