Rename Swift 1.5 fast variant to INT4 heads

QUANTIZATION_MANIFEST.json

@@ -5,7 +5,7 @@
     "revision": "9dba8a05877150d587215a519ce6befec3c9978a",
     "downloaded_at": 1790420447.6224113
   },
-  "variant": "fast",
+  "variant": "INT4 heads",
   "preflight_only": false,
   "unchanged_body_tensors_verified": 2382,
   "body": "unchanged asymmetric AWQ INT4 group128",
@@ -138,5 +138,6 @@
     "drafter/capture.py": "6bb7cff9e235999923fa83b994034be6c9246b45fa7d89cca6fabf067f030e9b",
     "drafter/train_mtp.py": "900951feef2e0ff7d729f9baac02e9125b5ec16de5680009e368e20139b79faf",
     "prepare/build_draft_vocab.py": "426f371a8db9f465c48b56ae05cbc782f0613980005405deac2ad95504d088bc"
-  }
+  },
+  "original_variant": "fast"
 }

README.md

@@ -16,7 +16,7 @@ tags:
 
 Model card written by GPT-6 Astra, refined with human feedback: 
 
-# Swift 1.5 HyperQwen-fast
+# Swift 1.5 HyperQwen — INT4 heads
 
 Swift 1.5's efficient reasoning, combined with HyperQwen's optimized serving, **delivered 37% lower average task completion time** than HyperQwen serving the Qwen fast checkpoint in our RTX 3090 evaluation.
 
@@ -36,7 +36,7 @@ Upstream checkpoint: [ukisai/Swift-1.5-Qwen3.8-27b-W4A16-AWQ](https://huggingfac
 
 ## Performance
 
-| Measurement | Qwen — HyperQwen fast quant (W4A16 AutoRound) | Swift 1.0 | Swift 1.5 INT8 heads | Swift 1.5 fast |
+| Measurement | Qwen — HyperQwen fast quant (W4A16 AutoRound) | Swift 1.0 | Swift 1.5 INT8 heads | Swift 1.5 INT4 heads |
 |---|---:|---:|---:|---:|
 | **Average request time ↓** | **108.1 s** | **66.2 s** | **72.2 s** | **68.2 s** |
 | **Average output tokens/task ↓** | **8,985** | **5,245** | **5,751** | **5,669** |
@@ -45,7 +45,7 @@ Upstream checkpoint: [ukisai/Swift-1.5-Qwen3.8-27b-W4A16-AWQ](https://huggingfac
 
 ## Quality
 
-| Test | Qwen — HyperQwen fast quant (W4A16 AutoRound) | Swift 1.0 | Swift 1.5 INT8 heads | Swift 1.5 fast |
+| Test | Qwen — HyperQwen fast quant (W4A16 AutoRound) | Swift 1.0 | Swift 1.5 INT8 heads | Swift 1.5 INT4 heads |
 |---|---:|---:|---:|---:|
 | **GSM8K — 200-question subset** | 97.5% | 98.0% | 98.0% | 97.5% |
 | **IFBench — 300 prompts, strict** | 74.0% | 73.3% | 73.7% | 72.3% |
@@ -60,7 +60,7 @@ Upstream checkpoint: [ukisai/Swift-1.5-Qwen3.8-27b-W4A16-AWQ](https://huggingfac
 - **Tool-call/JSON checks:** 20 custom weather-tool tasks checking the function name, arguments, Celsius-to-Fahrenheit conversion and final JSON; plus 10 JSON inventory-filtering tasks. These are integration checks, not an external agent benchmark.
 - **Perplexity:** 18,729 scored tokens from English Wikipedia and Python source; lower is better. Recomputed from saved token log-probabilities, excluding Danish; see [subset results](evaluation/perplexity-english-python.json).
 
-Compared with the Swift 1.5 INT8-head variant, this fast conversion measured **3.2% higher decode TPS** and **5.5% lower average request time**; benchmark score changes were mixed.
+Compared with the Swift 1.5 INT8-head variant, this INT4-head conversion measured **3.2% higher decode TPS** and **5.5% lower average request time**; benchmark score changes were mixed.
 
 **Evaluation setup:** RTX 3090 24 GB; FP8 KV cache; 150,000-token configured context; 128,000 output tokens per call. These are runtime settings, not fixed model properties, and the TPS test uses short prompts. Task evaluation used two concurrent requests; TPS was measured with one request at a time. All models used the same serving settings and task budgets. Thinking tests used xhigh effort, temperature 1.0, top_p 0.95, top_k 20 and seed 15027; GSM8K/tool checks were greedy.
 
@@ -69,8 +69,8 @@ Compared with the Swift 1.5 INT8-head variant, this fast conversion measured **3
 Requires the **patched HyperQwen runtime**, not stock vLLM or GGUF tools. From an installed HyperQwen checkout:
 
 ```bash
-hf download daavidhauser/Swift-1.5-Qwen3.8-27B-W4A16-HyperQwen-fast --local-dir models/Swift-1.5-Qwen3.8-27B-W4A16-HyperQwen-fast
-MODEL="$PWD/models/Swift-1.5-Qwen3.8-27B-W4A16-HyperQwen-fast" CTX=long MAX_LEN=150000 SPEC=mtp \
+hf download daavidhauser/Swift-1.5-Qwen3.8-27B-W4A16-HyperQwen-INT4 --local-dir models/Swift-1.5-Qwen3.8-27B-W4A16-HyperQwen-INT4
+MODEL="$PWD/models/Swift-1.5-Qwen3.8-27B-W4A16-HyperQwen-INT4" CTX=long MAX_LEN=150000 SPEC=mtp \
   bash single-user/start_qwen.sh
 ```
 

RELEASE_MANIFEST.json

@@ -1,5 +1,5 @@
 {
-  "repo": "daavidhauser/Swift-1.5-Qwen3.8-27B-W4A16-HyperQwen-fast",
+  "repo": "daavidhauser/Swift-1.5-Qwen3.8-27B-W4A16-HyperQwen-INT4",
   "files": {
     "LICENSE": {
       "size": 13306,
@@ -14,16 +14,16 @@
       "sha256": "e0e788911888bdcc087d34fc12dc5bf3818034762334de3575ea34d9559b186f"
     },
     "QUANTIZATION_MANIFEST.json": {
-      "size": 8398,
-      "sha256": "895d4e0dd27f896ed2a3a0e2160c59967cd10e8506f690b6a85c5cf90c529cec"
+      "size": 8434,
+      "sha256": "40ebbba877dd55e1c512e68785f3127673d57ac799c5ea7139761b563ed7de4b"
     },
     "README.md": {
-      "size": 5593,
-      "sha256": "100dec87d8127bb5d22cbb467c8442a24ef070c852b9089ce51fffb09435c7df"
+      "size": 5620,
+      "sha256": "de13ff8b19a75a4d6dcda0c9a36e276d3a5b19b74a44ef012bbfe39562f24a85"
     },
     "RUNTIME.md": {
       "size": 2608,
-      "sha256": "07b3d0091fee7fb8edf752fa2f3bb9118afe514eaba86291b88796aac0f906f5"
+      "sha256": "4a6a153fdf00ece6ddd976ee745951db61c5a36a69ba06eaf102a1ec1f5c3c74"
     },
     "chat_template.jinja": {
       "size": 8952,

RUNTIME.md

@@ -15,13 +15,13 @@ An independent clean-machine installation has not been tested.
 ## Download and launch from an installed HyperQwen checkout
 
 ```bash
-hf download daavidhauser/Swift-1.5-Qwen3.8-27B-W4A16-HyperQwen-fast \
-  --local-dir models/Swift-1.5-Qwen3.8-27B-W4A16-HyperQwen-fast
+hf download daavidhauser/Swift-1.5-Qwen3.8-27B-W4A16-HyperQwen-INT4 \
+  --local-dir models/Swift-1.5-Qwen3.8-27B-W4A16-HyperQwen-INT4
 
 # In the HyperQwen checkout; use the launcher snapshot accompanying the model.
-cp models/Swift-1.5-Qwen3.8-27B-W4A16-HyperQwen-fast/runtime/single-user/start_qwen.sh single-user/start_qwen.sh
+cp models/Swift-1.5-Qwen3.8-27B-W4A16-HyperQwen-INT4/runtime/single-user/start_qwen.sh single-user/start_qwen.sh
 
-MODEL="$PWD/models/Swift-1.5-Qwen3.8-27B-W4A16-HyperQwen-fast" \
+MODEL="$PWD/models/Swift-1.5-Qwen3.8-27B-W4A16-HyperQwen-INT4" \
 CTX=long MAX_LEN=150000 SPEC=mtp DRAFT_TOKENS=3 \
 PREFIX_CACHE=1 TOOLS=1 VISION=1 VISION_OFFLOAD=1 \
 GPU_UTIL=0.93 MAX_SEQS=8 API_SERVERS=1 HOST=127.0.0.1 PORT=18020 \