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": 77212.3208908301,
-      "required_pages_per_second_upper_estimate": 0.005341727516495148
+      "required_token_positions_per_second": 77043.49857025359,
+      "required_pages_per_second_upper_estimate": 0.005330047996635912
     },
     {
       "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": 9651.540111353763,
-      "required_pages_per_second_upper_estimate": 0.0040408982975476135
+      "required_token_positions_per_second": 9630.437321281699,
+      "required_pages_per_second_upper_estimate": 0.004032063000020808
     },
     {
       "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": 19303.080222707526,
-      "required_pages_per_second_upper_estimate": 0.004020322553931941
+      "required_token_positions_per_second": 19260.874642563398,
+      "required_pages_per_second_upper_estimate": 0.004011532244623916
     },
     {
       "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": 28954.62033406129,
-      "required_pages_per_second_upper_estimate": 0.0017357852778816325
+      "required_token_positions_per_second": 28891.3119638451,
+      "required_pages_per_second_upper_estimate": 0.001731990037753456
     },
     {
       "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": 28954.62033406129,
-      "required_pages_per_second_upper_estimate": 0.002251094373058015
+      "required_token_positions_per_second": 28891.3119638451,
+      "required_pages_per_second_upper_estimate": 0.0022461724257377977
     },
     {
       "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": 28954.62033406129,
-      "required_pages_per_second_upper_estimate": 0.37139878059621206
+      "required_token_positions_per_second": 28891.3119638451,
+      "required_pages_per_second_upper_estimate": 0.37058672879831067
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.1261457473750056,
-    "producer_p95_seconds": 0.27571753400115995,
-    "gpu_update_seconds": 0.9548735117578377,
-    "producer_batches_per_second": 7.927338184673033,
-    "gpu_batches_per_second": 1.0472591266659899,
-    "producer_headroom": 7.569605251290741,
+    "producer_batch_seconds": 0.09474884971862707,
+    "producer_p95_seconds": 0.27294528299898957,
+    "gpu_update_seconds": 0.9569658876896628,
+    "producer_batches_per_second": 10.55421783979089,
+    "gpu_batches_per_second": 1.0449693273960177,
+    "producer_headroom": 10.100026443925566,
     "sustainable": true,
-    "buffer_seconds": 60.15703124074378,
+    "buffer_seconds": 60.28885092444876,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 452654.8201504785,
-    "upload_bytes_per_second": 6032477.764523069,
+    "generation_bytes_per_second": 502144.5564158596,
+    "upload_bytes_per_second": 5944201.831218916,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.12918713937536,
+    "estimated_full_shards_per_hour": 107.74853247982827,
     "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": 0.9355267249993631,
-      "bytes": 7522148,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.2877170510000724,
       "bytes": 6767236,
@@ -562,6 +556,12 @@
       "bytes": 12336426,
       "pages": 42,
       "shards": 4
+    },
+    {
+      "seconds": 0.8797184569993988,
+      "bytes": 870056,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -264,3 +264,4 @@
 {"step": 423500, "loss": 0.5925978422164917, "lr": 0.0003, "grad_norm": 0.1639069225832975, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61988663011, "tokens_per_second": 152502.25051917668, "data_wait_s": 1.3473996659740806e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 423600, "loss": 0.5602745413780212, "lr": 0.0003, "grad_norm": 0.16093032548225664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62003320787, "tokens_per_second": 151527.86366619193, "data_wait_s": 1.3237000530352816e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 423700, "loss": 0.5932236909866333, "lr": 0.0003, "grad_norm": 0.17331497094262296, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62017961215, "tokens_per_second": 152020.00371867706, "data_wait_s": 2.2943000658415258e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 423800, "loss": 0.5542709231376648, "lr": 0.0003, "grad_norm": 0.13283663005110638, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62032678547, "tokens_per_second": 152169.60309610044, "data_wait_s": 1.1085998266935349e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:00e17f3007f38ee8f166d358f0ad3c64e7cfc7d01417bd5e12a237cc299c510e
-size 567468
+oid sha256:ecbde87896bfc8038b42f71701aace408397fd04d227ffe78843b4e886a5227c
+size 569586