Layer-pruned Ovis2-4B (PPL-based (Shortened LLaMA), removed 10 layers, 17.9% reduction)

config.json

@@ -111,7 +111,7 @@
   },
   "model_type": "ovis",
   "multimodal_max_length": 32768,
-  "num_hidden_layers": 18,
+  "num_hidden_layers": 26,
   "transformers_version": "5.3.0",
   "visual_tokenizer_config": {
     "_attn_implementation_autoset": true,

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:327f7dc09cc11d1971ae7aa31aa6de5f27d9cb490a293b283409f5c989adaf05
-size 6457490620
+oid sha256:f3a5829c5ba683148706c7226f8dd1b8abfb132bc572cf436f77430e4843d5bb
+size 7690733916

pruning_info.json

@@ -9,44 +9,36 @@
     6,
     7,
     8,
-    9,
     10,
     11,
-    12,
-    13,
-    14,
-    15,
-    16,
     18,
-    19,
-    20,
-    22
+    19
   ],
   "n_layers_original": 36,
-  "n_layers_remaining": 18,
+  "n_layers_remaining": 26,
   "params_before_M": 4304.9,
-  "params_after_M": 2917.6,
-  "param_reduction_pct": 32.2,
+  "params_after_M": 3534.2,
+  "param_reduction_pct": 17.9,
   "benchmarks": {
     "vqav2": {
-      "accuracy": 0.0,
-      "avg_latency_s": 0.6058,
-      "peak_memory_mb": 15225.1,
-      "avg_memory_mb": 15222.8,
-      "throughput_sps": 1.65,
-      "avg_power_w": 226.7,
-      "avg_gpu_util_pct": 64.3,
+      "accuracy": 0.06,
+      "avg_latency_s": 0.5029,
+      "peak_memory_mb": 15235.1,
+      "avg_memory_mb": 15230.8,
+      "throughput_sps": 1.987,
+      "avg_power_w": 236.7,
+      "avg_gpu_util_pct": 69.1,
       "n_samples": 50,
       "n_evaluated": 50,
       "n_skipped": 0,
       "all_failed": false,
-      "zero_accuracy_warning": true,
+      "zero_accuracy_warning": false,
       "metrics": {
-        "exact_match": 0.0,
-        "contains": 0.1,
-        "token_f1": 0.0022,
-        "bleu": 0.0012,
-        "rouge_l": 0.0022
+        "exact_match": 0.06,
+        "contains": 0.42,
+        "token_f1": 0.1762,
+        "bleu": 0.135,
+        "rouge_l": 0.1762
       }
     }
   }