Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74724.08890194965,
-      "required_pages_per_second_upper_estimate": 0.006476894639549129
+      "required_token_positions_per_second": 74511.34786027142,
+      "required_pages_per_second_upper_estimate": 0.0064584547852437005
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9340.511112743707,
-      "required_pages_per_second_upper_estimate": 0.0037260695359596723
+      "required_token_positions_per_second": 9313.918482533927,
+      "required_pages_per_second_upper_estimate": 0.0037154613381737383
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18681.022225487413,
-      "required_pages_per_second_upper_estimate": 0.004321525717104397
+      "required_token_positions_per_second": 18627.836965067854,
+      "required_pages_per_second_upper_estimate": 0.004309222243134946
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28021.53333823112,
-      "required_pages_per_second_upper_estimate": 0.00170636436267497
+      "required_token_positions_per_second": 27941.75544760178,
+      "required_pages_per_second_upper_estimate": 0.0017015063076978871
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28021.53333823112,
-      "required_pages_per_second_upper_estimate": 0.002157175204945921
+      "required_token_positions_per_second": 27941.75544760178,
+      "required_pages_per_second_upper_estimate": 0.002151033682085939
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28021.53333823112,
-      "required_pages_per_second_upper_estimate": 0.34793366202156917
+      "required_token_positions_per_second": 27941.75544760178,
+      "required_pages_per_second_upper_estimate": 0.3469430876193772
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11200499768705185,
-    "producer_p95_seconds": 0.3322004139990895,
-    "gpu_update_seconds": 0.9866697752145672,
-    "producer_batches_per_second": 8.928173033796718,
-    "gpu_batches_per_second": 1.0135103203931972,
-    "producer_headroom": 8.809158480332966,
+    "producer_batch_seconds": 0.10234349722657043,
+    "producer_p95_seconds": 0.3404590539976198,
+    "gpu_update_seconds": 0.9894868649841042,
+    "producer_batches_per_second": 9.771016499330452,
+    "gpu_batches_per_second": 1.010624835344393,
+    "producer_headroom": 9.668292483630445,
     "sustainable": true,
-    "buffer_seconds": 60.1868562880886,
+    "buffer_seconds": 60.35869876403036,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 409745.259217941,
-    "upload_bytes_per_second": 2878932.466301815,
+    "generation_bytes_per_second": 424331.3598140253,
+    "upload_bytes_per_second": 2874349.8796414747,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.92179424670861,
+    "estimated_full_shards_per_hour": 91.05163188758439,
     "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": 1.104966183000215,
-      "bytes": 6761570,
-      "pages": 43,
-      "shards": 3
-    },
     {
       "seconds": 1.8051689160020032,
       "bytes": 11661330,
@@ -562,6 +556,12 @@
       "bytes": 6951757,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.125239683999098,
+      "bytes": 6131144,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -243,3 +243,4 @@
 {"step": 219300, "loss": 0.5816761590540409, "lr": 0.0003, "grad_norm": 0.14476674569779652, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32098461872, "tokens_per_second": 148982.575581119, "data_wait_s": 1.609200262464583e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 219400, "loss": 0.5835527759045362, "lr": 0.0003, "grad_norm": 0.15944790522468322, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32113115322, "tokens_per_second": 148725.52193367132, "data_wait_s": 1.5837998944334686e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 219500, "loss": 0.5853289980441332, "lr": 0.0003, "grad_norm": 0.1496752562692594, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32127750282, "tokens_per_second": 148218.00638480662, "data_wait_s": 1.4034998457645997e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 219600, "loss": 0.5990609619766474, "lr": 0.0003, "grad_norm": 0.1499999727558524, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 32142426748, "tokens_per_second": 149850.29394667826, "data_wait_s": 1.529600194771774e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4422b17dd9467e83ca7c4ba5b8bdc098bac745227faec10bf99930b02bf6567d
-size 522207
+oid sha256:d4034aaebab22f063333f2041cdd380eb60c0a8d110658f8e2d5ad7ca67b26b2
+size 524325