Upload folder using huggingface_hub

README.md

@@ -56,21 +56,6 @@ The report contains one row for each (Dataset × Context) pair. Values are avera
 | general_qa | 1k | 1039 | 221 | 463.2 | 2.59 | 16.4 | 15.79 |
 | general_qa | 4k | 3978 | 243 | 496.4 | 8.31 | 15.0 | 24.22 |
 
-## Measured results (reference)
-
-Measured deployment: `http://10.77.53.211:8080`, GGUF **Q8_0**, `n_ctx=131072`, llama.cpp fingerprint `b11103-a78cbbf70` (see `config.json`).
-
-| Context | Prefill tok/s | Decode tok/s | TTFT | Total time per request |
-|---|---|---|---|---|
-| 1K | 509.5 | 17.9 | 2.3s | 16.3s |
-| 4K | 548.9 | 16.1 | 7.8s | 23.4s |
-| 64K | 300.5 | **5.9** | 218.6s | 261.2s |
-| 128K | — | — | — | **Server crashed during the request** (see below) |
-
-The difference between the two repeated runs at each tier was less than 0.3%, indicating good measurement repeatability for this endpoint.
-
-**Known issue:** After the 128K request (130,667 prompt tokens) was sent, the server died during prefill. The client waited 240 seconds before the connection was closed, after which the port returned `Connection refused`. The deployment's practical limit is below its declared `n_ctx`; confirm that the available GPU and system memory can hold the corresponding KV cache before running the full 128K tier.
-
 ## Dataset mode: `datasets/`
 
 The benchmark uses real-world text as input, which better represents practical workloads than synthetic repeated text. The dataset contains **2 families × 3 tiers × 10 samples = 60 samples**: