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": 77003.96081171921,
-      "required_pages_per_second_upper_estimate": 0.005327312682760265
+      "required_token_positions_per_second": 75598.86968125336,
+      "required_pages_per_second_upper_estimate": 0.005230105218094038
     },
     {
       "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": 9625.495101464901,
-      "required_pages_per_second_upper_estimate": 0.0040299937957888
+      "required_token_positions_per_second": 9449.85871015667,
+      "required_pages_per_second_upper_estimate": 0.003956458506452963
     },
     {
       "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": 19250.990202929803,
-      "required_pages_per_second_upper_estimate": 0.00400947357651844
+      "required_token_positions_per_second": 18899.71742031334,
+      "required_pages_per_second_upper_estimate": 0.0039363127195856645
     },
     {
       "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": 28876.485304394708,
-      "required_pages_per_second_upper_estimate": 0.0017311012021584034
+      "required_token_positions_per_second": 28349.576130470006,
+      "required_pages_per_second_upper_estimate": 0.0016995138017253414
     },
     {
       "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": 28876.485304394708,
-      "required_pages_per_second_upper_estimate": 0.0022450197181811127
+      "required_token_positions_per_second": 28349.576130470006,
+      "required_pages_per_second_upper_estimate": 0.002204054847536989
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28876.485304394708,
-      "required_pages_per_second_upper_estimate": 0.37039654833050767
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28349.576130470006,
+      "required_pages_per_second_upper_estimate": 0.36363792319839416
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10498235464851291,
-    "producer_p95_seconds": 0.18780492899986712,
-    "gpu_update_seconds": 0.9574572427549642,
-    "producer_batches_per_second": 9.525410278213503,
-    "gpu_batches_per_second": 1.0444330622249243,
-    "producer_headroom": 9.120173061088098,
+    "producer_batch_seconds": 0.23268139603123217,
+    "producer_p95_seconds": 0.18061906499951874,
+    "gpu_update_seconds": 0.9752526765394578,
+    "producer_batches_per_second": 4.297722194626049,
+    "gpu_batches_per_second": 1.0253752940708192,
+    "producer_headroom": 4.191365073332087,
     "sustainable": true,
-    "buffer_seconds": 60.31980629356275,
+    "buffer_seconds": 60.46566594544638,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 458694.353659499,
-    "upload_bytes_per_second": 5437986.444895973,
+    "generation_bytes_per_second": 438343.8641476351,
+    "upload_bytes_per_second": 5433148.26919615,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.4251304372666,
+    "estimated_full_shards_per_hour": 94.05838912317076,
     "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": [
@@ -502,6 +502,12 @@
       "bytes": 5051210,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.4254050959998494,
+      "bytes": 7441828,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -63,3 +63,4 @@
 {"step": 610700, "loss": 0.5271191000938416, "lr": 0.0001, "grad_norm": 0.1514641595808708, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89388852771, "tokens_per_second": 151549.9499952516, "data_wait_s": 1.0144999578187708e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 610800, "loss": 0.5893732309341431, "lr": 0.0001, "grad_norm": 0.15391375382795708, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89403323012, "tokens_per_second": 150141.51985255766, "data_wait_s": 1.0426999324408825e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 610900, "loss": 0.5411556363105774, "lr": 0.0001, "grad_norm": 0.1468534848009401, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89417853373, "tokens_per_second": 150565.84804299244, "data_wait_s": 1.4136000572761986e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 611000, "loss": 0.5577870011329651, "lr": 0.0001, "grad_norm": 0.15734135282543868, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89432608088, "tokens_per_second": 153349.19443355923, "data_wait_s": 1.5098999938345514e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:98defb4bc72f5984b0ee137cc782fc0ddfea129be8cca80d3844c9fbbc6fb6d5
-size 139401
+oid sha256:0bf853712f1be7cbbf0016c363e02995357c4fb9b5d47ac98755a72518c4dafc
+size 141519