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": 76526.67883835998,
-      "required_pages_per_second_upper_estimate": 0.005294293208396518
+      "required_token_positions_per_second": 76869.40119144382,
+      "required_pages_per_second_upper_estimate": 0.005318003536008278
     },
     {
       "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": 9565.834854794997,
-      "required_pages_per_second_upper_estimate": 0.004005015296355732
+      "required_token_positions_per_second": 9608.675148930477,
+      "required_pages_per_second_upper_estimate": 0.004022951632903187
     },
     {
       "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": 19131.669709589994,
-      "required_pages_per_second_upper_estimate": 0.003984622264448774
+      "required_token_positions_per_second": 19217.350297860954,
+      "required_pages_per_second_upper_estimate": 0.004002467271436554
     },
     {
       "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": 28697.504564384995,
-      "required_pages_per_second_upper_estimate": 0.001720371580082587
+      "required_token_positions_per_second": 28826.02544679143,
+      "required_pages_per_second_upper_estimate": 0.0017280762107428282
     },
     {
       "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": 28697.504564384995,
-      "required_pages_per_second_upper_estimate": 0.0022311047528984335
+      "required_token_positions_per_second": 28826.02544679143,
+      "required_pages_per_second_upper_estimate": 0.0022410966862019153
     },
     {
       "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": 28697.504564384995,
-      "required_pages_per_second_upper_estimate": 0.3681007755722091
+      "required_token_positions_per_second": 28826.02544679143,
+      "required_pages_per_second_upper_estimate": 0.3697493034567467
     }
   ],
   "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": 0.21477461474231063,
-    "producer_p95_seconds": 0.28193192400067346,
-    "gpu_update_seconds": 0.9634287168757008,
-    "producer_batches_per_second": 4.656043737756499,
-    "gpu_batches_per_second": 1.0379595111539712,
-    "producer_headroom": 4.4857662439838855,
+    "producer_batch_seconds": 0.10607404214857752,
+    "producer_p95_seconds": 0.2764453179988777,
+    "gpu_update_seconds": 0.9591332683388526,
+    "producer_batches_per_second": 9.427377139067667,
+    "gpu_batches_per_second": 1.0426079805697133,
+    "producer_headroom": 9.042111047256954,
     "sustainable": true,
-    "buffer_seconds": 60.696009163169144,
+    "buffer_seconds": 60.42539590534771,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 169,
+    "upload_pages_per_commit": 165,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 591236.8657698374,
-    "upload_bytes_per_second": 5277639.819981935,
+    "generation_bytes_per_second": 588017.2038507212,
+    "upload_bytes_per_second": 5327953.930747897,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.432834052187644,
+    "estimated_full_shards_per_hour": 63.087401803213254,
     "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": 1.401579060999211,
-      "bytes": 925271,
-      "pages": 28,
-      "shards": 1
-    },
     {
       "seconds": 1.7116284690018801,
       "bytes": 11540303,
@@ -562,6 +556,12 @@
       "bytes": 10304832,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 0.9087454849977803,
+      "bytes": 2381627,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -257,3 +257,4 @@
 {"step": 670700, "loss": 0.5527182817459106, "lr": 0.0001, "grad_norm": 0.19289207468098582, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98173372423, "tokens_per_second": 151383.34119689272, "data_wait_s": 9.703002433525398e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 670800, "loss": 0.5319862961769104, "lr": 0.0001, "grad_norm": 0.1798059597896993, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98187887326, "tokens_per_second": 150440.26937082288, "data_wait_s": 1.5022997104097158e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 670900, "loss": 0.5567914247512817, "lr": 0.0001, "grad_norm": 0.20301505027776218, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98202578576, "tokens_per_second": 152670.8285389809, "data_wait_s": 7.121998351067305e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 671000, "loss": 0.558301568031311, "lr": 0.0001, "grad_norm": 0.20087257593448588, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98217243841, "tokens_per_second": 152556.51921369898, "data_wait_s": 1.180199978989549e-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 sha256:41dbffa8088c7ee8bad2c1ca1ed2c7b35d19d0c08879479f24a0f603c0983274
-size 552642
+oid sha256:a139b2756396851d01a3f66c3af149a6f8edfef636e819826f94e0ca0528c782
+size 554760