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": 77292.3550304763,
-      "required_pages_per_second_upper_estimate": 0.005347264463980661
+      "required_token_positions_per_second": 76587.28254831888,
+      "required_pages_per_second_upper_estimate": 0.005298485913671456
     },
     {
       "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": 9661.544378809538,
-      "required_pages_per_second_upper_estimate": 0.004045086875418486
+      "required_token_positions_per_second": 9573.41031853986,
+      "required_pages_per_second_upper_estimate": 0.004008186984832034
     },
     {
       "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": 19323.088757619076,
-      "required_pages_per_second_upper_estimate": 0.004024489804093467
+      "required_token_positions_per_second": 19146.82063707972,
+      "required_pages_per_second_upper_estimate": 0.003987777803088057
     },
     {
       "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": 28984.633136428616,
-      "required_pages_per_second_upper_estimate": 0.0017375844995566581
+      "required_token_positions_per_second": 28720.230955619576,
+      "required_pages_per_second_upper_estimate": 0.00172173399253591
     },
     {
       "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": 28984.633136428616,
-      "required_pages_per_second_upper_estimate": 0.0022534277364297086
+      "required_token_positions_per_second": 28720.230955619576,
+      "required_pages_per_second_upper_estimate": 0.0022328716298540916
     },
     {
       "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": 28984.633136428616,
-      "required_pages_per_second_upper_estimate": 0.37178375259974367
+      "required_token_positions_per_second": 28720.230955619576,
+      "required_pages_per_second_upper_estimate": 0.3683922853172686
     }
   ],
   "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": 0.08076144157018916,
-    "producer_p95_seconds": 0.2235941129983985,
-    "gpu_update_seconds": 0.9538847661056402,
-    "producer_batches_per_second": 12.382146486710585,
-    "gpu_batches_per_second": 1.048344659159021,
-    "producer_headroom": 11.811140905361702,
+    "producer_batch_seconds": 0.278069025906305,
+    "producer_p95_seconds": 0.22689579699726892,
+    "gpu_update_seconds": 0.9626663533006938,
+    "producer_batches_per_second": 3.5962293777263374,
+    "gpu_batches_per_second": 1.0387815015776758,
+    "producer_headroom": 3.4619690206886364,
     "sustainable": true,
-    "buffer_seconds": 60.094740264655336,
+    "buffer_seconds": 60.647980257943715,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 456073.34529352654,
-    "upload_bytes_per_second": 1886866.1050289606,
+    "generation_bytes_per_second": 418029.72941294033,
+    "upload_bytes_per_second": 1891832.1223800185,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.86272305588099,
+    "estimated_full_shards_per_hour": 89.6994486979595,
     "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": 1.8339393939986621,
-      "bytes": 7623870,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.1467835090006702,
       "bytes": 6065504,
@@ -562,6 +556,12 @@
       "bytes": 11929043,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.140193093000562,
+      "bytes": 7462921,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -163,3 +163,4 @@
 {"step": 661300, "loss": 0.501309871673584, "lr": 0.0001, "grad_norm": 0.1910924292280799, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96796881185, "tokens_per_second": 153321.34276279656, "data_wait_s": 1.4727000234415755e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 661400, "loss": 0.5986628532409668, "lr": 0.0001, "grad_norm": 0.20128587924484945, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96811615482, "tokens_per_second": 154072.87834906369, "data_wait_s": 8.910001270123757e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 661500, "loss": 0.5121641159057617, "lr": 0.0001, "grad_norm": 0.19666728712620124, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96826378579, "tokens_per_second": 153557.48604135014, "data_wait_s": 9.351999324280769e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 661600, "loss": 0.5818714499473572, "lr": 0.0001, "grad_norm": 0.18780524546508917, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96841124320, "tokens_per_second": 153551.00608325418, "data_wait_s": 1.268800042453222e-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 sha256:85d2f3dea8071fc4214805e362cf0f8f43536f7def58cda53b818c6be617011a
-size 351984
+oid sha256:2d2d0afda716ce59deef7923cca9d3ce90ac3a6f3323fad6b30dda3841cc6ca9
+size 354102