Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77018.9329996887,
-      "required_pages_per_second_upper_estimate": 0.005328348493464256
+      "required_token_positions_per_second": 76528.66632985527,
+      "required_pages_per_second_upper_estimate": 0.005294430707669779
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9627.366624961087,
-      "required_pages_per_second_upper_estimate": 0.0040307773635947985
+      "required_token_positions_per_second": 9566.08329123191,
+      "required_pages_per_second_upper_estimate": 0.004005119311503925
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19254.733249922174,
-      "required_pages_per_second_upper_estimate": 0.004010253154496164
+      "required_token_positions_per_second": 19132.16658246382,
+      "required_pages_per_second_upper_estimate": 0.003984725749964972
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28882.099874883257,
-      "required_pages_per_second_upper_estimate": 0.001731437787086239
+      "required_token_positions_per_second": 28698.24987369573,
+      "required_pages_per_second_upper_estimate": 0.0017204162602377443
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28882.099874883257,
-      "required_pages_per_second_upper_estimate": 0.0022454562263407116
+      "required_token_positions_per_second": 28698.24987369573,
+      "required_pages_per_second_upper_estimate": 0.002231162697419074
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28882.099874883257,
-      "required_pages_per_second_upper_estimate": 0.3704685660122787
+      "required_token_positions_per_second": 28698.24987369573,
+      "required_pages_per_second_upper_estimate": 0.36811033559979645
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.40654050267184516,
-    "producer_p95_seconds": 0.258296321000671,
-    "gpu_update_seconds": 0.9572711167044863,
-    "producer_batches_per_second": 2.4597795137946896,
-    "gpu_batches_per_second": 1.0446361355209512,
-    "producer_headroom": 2.354675882017061,
+    "producer_batch_seconds": 0.06603346498448559,
+    "producer_p95_seconds": 0.22812866100139217,
+    "gpu_update_seconds": 0.9634036961027937,
+    "producer_batches_per_second": 15.14383654159217,
+    "gpu_batches_per_second": 1.0379864682326292,
+    "producer_headroom": 14.589628097346447,
     "sustainable": true,
-    "buffer_seconds": 60.30808035238264,
+    "buffer_seconds": 60.694432854476005,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 344202.33776186995,
-    "upload_bytes_per_second": 5111897.279630143,
+    "generation_bytes_per_second": 343841.1233995657,
+    "upload_bytes_per_second": 5099116.791732661,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.85780906335901,
+    "estimated_full_shards_per_hour": 73.78030086984852,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.2133236099998612,
-      "bytes": 11087663,
-      "pages": 9,
-      "shards": 3
-    },
     {
       "seconds": 1.060693833998812,
       "bytes": 7602919,
@@ -562,6 +556,12 @@
       "bytes": 2686677,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.3082805390004069,
+      "bytes": 10490550,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -179,3 +179,4 @@
 {"step": 777900, "loss": 0.549085795879364, "lr": 3e-05, "grad_norm": 0.24472834686400965, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113862549260, "tokens_per_second": 151264.83046431563, "data_wait_s": 1.7368998669553548e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 778000, "loss": 0.5105672478675842, "lr": 3e-05, "grad_norm": 0.2628503965985092, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113877148301, "tokens_per_second": 151913.99886996634, "data_wait_s": 7.556001946795732e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 778100, "loss": 0.511470377445221, "lr": 3e-05, "grad_norm": 0.2513624045672413, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113891846413, "tokens_per_second": 152039.9511622395, "data_wait_s": 1.0210998880211264e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 778200, "loss": 0.5231640338897705, "lr": 3e-05, "grad_norm": 0.24825772797739806, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113906537906, "tokens_per_second": 152285.53860219131, "data_wait_s": 1.3607001164928079e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cbf9b867a28accfa0c9cb77c7a8bbe1a4a1761f4279f462a811a7e8192b2fdaa
-size 385872
+oid sha256:985bee127869bf916734b03a77afc056d9c86d7c70866fa5b31ca041de6022dd
+size 387990