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": 75629.22489837605,
-      "required_pages_per_second_upper_estimate": 0.00655534953367316
+      "required_token_positions_per_second": 75683.42606068122,
+      "required_pages_per_second_upper_estimate": 0.0065600475530501814
     },
     {
       "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": 9453.653112297006,
-      "required_pages_per_second_upper_estimate": 0.0037712035712051247
+      "required_token_positions_per_second": 9460.428257585152,
+      "required_pages_per_second_upper_estimate": 0.003773906277958015
     },
     {
       "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": 18907.306224594013,
-      "required_pages_per_second_upper_estimate": 0.004373872537835342
+      "required_token_positions_per_second": 18920.856515170304,
+      "required_pages_per_second_upper_estimate": 0.0043770071590831985
     },
     {
       "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": 28360.959336891017,
-      "required_pages_per_second_upper_estimate": 0.0017270336251627727
+      "required_token_positions_per_second": 28381.284772755454,
+      "required_pages_per_second_upper_estimate": 0.0017282713375675965
     },
     {
       "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": 28360.959336891017,
-      "required_pages_per_second_upper_estimate": 0.0021833051579140607
+      "required_token_positions_per_second": 28381.284772755454,
+      "required_pages_per_second_upper_estimate": 0.0021848698662313127
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28360.959336891017,
-      "required_pages_per_second_upper_estimate": 0.3521481969391834
+      "required_token_positions_per_second": 28381.284772755454,
+      "required_pages_per_second_upper_estimate": 0.3524005708277618
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.14163054111702422,
-    "producer_p95_seconds": 0.40006909200019436,
-    "gpu_update_seconds": 0.9748612404671508,
-    "producer_batches_per_second": 7.06062401592984,
-    "gpu_batches_per_second": 1.0257870130530604,
-    "producer_headroom": 6.883128686641521,
+    "producer_batch_seconds": 0.0994280640704801,
+    "producer_p95_seconds": 0.40244642600009684,
+    "gpu_update_seconds": 0.9741630874491147,
+    "producer_batches_per_second": 10.057522585284822,
+    "gpu_batches_per_second": 1.0265221633664443,
+    "producer_headroom": 9.797667253770264,
     "sustainable": true,
-    "buffer_seconds": 60.441396908963355,
+    "buffer_seconds": 60.398111421845115,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 280694.9900838151,
-    "upload_bytes_per_second": 5859894.418154995,
+    "generation_bytes_per_second": 280909.8213823769,
+    "upload_bytes_per_second": 5826216.335638164,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.23061062703934,
+    "estimated_full_shards_per_hour": 60.276708422694014,
     "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": 0.8879908719973173,
-      "bytes": 2645607,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.5463334099986241,
       "bytes": 15909077,
@@ -562,6 +556,12 @@
       "bytes": 13247311,
       "pages": 40,
       "shards": 4
+    },
+    {
+      "seconds": 1.0703849589990568,
+      "bytes": 1241496,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -392,3 +392,4 @@
 {"step": 234200, "loss": 0.6237741932272911, "lr": 0.0003, "grad_norm": 0.1550310669100969, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34280619420, "tokens_per_second": 136297.61284754198, "data_wait_s": 1.2418997357599437e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 234300, "loss": 0.5962309408932924, "lr": 0.0003, "grad_norm": 0.13739073586915035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34295097165, "tokens_per_second": 147068.08691130526, "data_wait_s": 8.19200067780912e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 234400, "loss": 0.6019500847905874, "lr": 0.0003, "grad_norm": 0.14625804607041817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34309682397, "tokens_per_second": 148176.22259378296, "data_wait_s": 9.702998795546591e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 234500, "loss": 0.6188864409923553, "lr": 0.0003, "grad_norm": 0.14668536728967518, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34324272343, "tokens_per_second": 147640.8186086836, "data_wait_s": 1.7001002561300993e-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 sha256:16dfe39fc9436345587028f21c200594c46464d83542b14525bde64bc73df876
-size 840138
+oid sha256:8dab535c52533c5de6abb6d4a6bdd6582ef272fddc3231b321338329f4e94d1a
+size 842256