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": 76639.76533724267,
-      "required_pages_per_second_upper_estimate": 0.005302116794786052
+      "required_token_positions_per_second": 77153.8868606227,
+      "required_pages_per_second_upper_estimate": 0.00533768491469977
     },
     {
       "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": 9579.970667155334,
-      "required_pages_per_second_upper_estimate": 0.004010933665801676
+      "required_token_positions_per_second": 9644.235857577838,
+      "required_pages_per_second_upper_estimate": 0.0040378401590217975
     },
     {
       "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": 19159.941334310668,
-      "required_pages_per_second_upper_estimate": 0.003990510498305208
+      "required_token_positions_per_second": 19288.471715155676,
+      "required_pages_per_second_upper_estimate": 0.0040172799870611415
     },
     {
       "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": 28739.912001466,
-      "required_pages_per_second_upper_estimate": 0.0017229138411831862
+      "required_token_positions_per_second": 28932.707572733514,
+      "required_pages_per_second_upper_estimate": 0.0017344716412988816
     },
     {
       "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": 28739.912001466,
-      "required_pages_per_second_upper_estimate": 0.002234401744600879
+      "required_token_positions_per_second": 28932.707572733514,
+      "required_pages_per_second_upper_estimate": 0.0022493907522488314
     },
     {
       "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": 28739.912001466,
-      "required_pages_per_second_upper_estimate": 0.3686447326415259
+      "required_token_positions_per_second": 28932.707572733514,
+      "required_pages_per_second_upper_estimate": 0.3711177072219894
     }
   ],
   "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": 0.2142487486950131,
-    "producer_p95_seconds": 0.3454566420041374,
-    "gpu_update_seconds": 0.9620071209191489,
-    "producer_batches_per_second": 4.667471833982647,
-    "gpu_batches_per_second": 1.039493344960431,
-    "producer_headroom": 4.490141140980866,
+    "producer_batch_seconds": 0.0894173764139623,
+    "producer_p95_seconds": 0.3180558839958394,
+    "gpu_update_seconds": 0.9555967041970094,
+    "producer_batches_per_second": 11.183508621081087,
+    "gpu_batches_per_second": 1.0464665644073174,
+    "producer_headroom": 10.686923979663927,
     "sustainable": true,
-    "buffer_seconds": 60.60644861790638,
+    "buffer_seconds": 60.20259236441159,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 404315.30007759656,
-    "upload_bytes_per_second": 5530254.389316219,
+    "generation_bytes_per_second": 404405.41966028657,
+    "upload_bytes_per_second": 5501910.391762454,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.7566514181702,
+    "estimated_full_shards_per_hour": 86.77598898273895,
     "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": 1.5010855370019272,
-      "bytes": 6022741,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0192723729996942,
       "bytes": 2390933,
@@ -562,6 +556,12 @@
       "bytes": 13416312,
       "pages": 42,
       "shards": 4
+    },
+    {
+      "seconds": 0.995916217994818,
+      "bytes": 1003302,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -342,3 +342,4 @@
 {"step": 679200, "loss": 0.6018109321594238, "lr": 0.0001, "grad_norm": 0.21745714831770144, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99417364559, "tokens_per_second": 150898.36260275112, "data_wait_s": 1.4440003724303097e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 679300, "loss": 0.5146587491035461, "lr": 0.0001, "grad_norm": 0.1973563229490574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99431945721, "tokens_per_second": 151347.31785791268, "data_wait_s": 1.2218995834700763e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 679400, "loss": 0.6518879532814026, "lr": 0.0001, "grad_norm": 0.2733330479010058, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99446621460, "tokens_per_second": 153069.31936941762, "data_wait_s": 1.0310999641660601e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 679500, "loss": 0.5391449332237244, "lr": 0.0001, "grad_norm": 0.19859551373639187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99461344676, "tokens_per_second": 152318.18885355955, "data_wait_s": 1.2577998859342188e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d84d9f1aa4e9bff91bca2275da42c8fb4f850e29bd963f254464815dd449349b
-size 733455
+oid sha256:3f9d4c3888ee1c81a159a2b1eccc4959cdea6e46178a58ccefe8f4c93cda1997
+size 735573