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": 76630.44645987086,
-      "required_pages_per_second_upper_estimate": 0.005301472093226695
+      "required_token_positions_per_second": 76912.37749335157,
+      "required_pages_per_second_upper_estimate": 0.005320976736293015
     },
     {
       "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": 9578.805807483857,
-      "required_pages_per_second_upper_estimate": 0.004010445963382264
+      "required_token_positions_per_second": 9614.047186668946,
+      "required_pages_per_second_upper_estimate": 0.004025200792923379
     },
     {
       "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": 19157.611614967715,
-      "required_pages_per_second_upper_estimate": 0.003990025279204902
+      "required_token_positions_per_second": 19228.09437333789,
+      "required_pages_per_second_upper_estimate": 0.004004704979018075
     },
     {
       "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": 28736.41742245157,
-      "required_pages_per_second_upper_estimate": 0.0017227043465071843
+      "required_token_positions_per_second": 28842.14156000684,
+      "required_pages_per_second_upper_estimate": 0.001729042347122212
     },
     {
       "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": 28736.41742245157,
-      "required_pages_per_second_upper_estimate": 0.002234130056453536
+      "required_token_positions_per_second": 28842.14156000684,
+      "required_pages_per_second_upper_estimate": 0.0022423496431171227
     },
     {
       "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": 28736.41742245157,
-      "required_pages_per_second_upper_estimate": 0.3685999079341154
+      "required_token_positions_per_second": 28842.14156000684,
+      "required_pages_per_second_upper_estimate": 0.3699560236529398
     }
   ],
   "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": 0.6137633864847771,
-    "producer_p95_seconds": 0.3269848319978337,
-    "gpu_update_seconds": 0.9621241087067034,
-    "producer_batches_per_second": 1.6292923657882654,
-    "gpu_batches_per_second": 1.0393669495967726,
-    "producer_headroom": 1.567581465256671,
+    "producer_batch_seconds": 0.1087129942421825,
+    "producer_p95_seconds": 0.32735338999918895,
+    "gpu_update_seconds": 0.9585973337825003,
+    "producer_batches_per_second": 9.198532401492653,
+    "gpu_batches_per_second": 1.0431908839701547,
+    "producer_headroom": 8.817688634782797,
     "sustainable": true,
-    "buffer_seconds": 60.613818848522314,
+    "buffer_seconds": 60.39163202829752,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 247554.08361550872,
-    "upload_bytes_per_second": 4528909.269040037,
+    "generation_bytes_per_second": 279524.8772814654,
+    "upload_bytes_per_second": 4630818.231819533,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.11934357975908,
+    "estimated_full_shards_per_hour": 59.97953165848705,
     "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": 1.0437422140021226,
-      "bytes": 887331,
-      "pages": 27,
-      "shards": 1
-    },
     {
       "seconds": 1.3929757290025009,
       "bytes": 11170736,
@@ -562,6 +556,12 @@
       "bytes": 1790342,
       "pages": 13,
       "shards": 2
+    },
+    {
+      "seconds": 1.4330027210016851,
+      "bytes": 12099864,
+      "pages": 47,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -410,3 +410,4 @@
 {"step": 761000, "loss": 0.5758531093597412, "lr": 5e-05, "grad_norm": 0.2517351415603714, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111387964808, "tokens_per_second": 149993.37426602026, "data_wait_s": 1.098900247598067e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 761100, "loss": 0.5433309078216553, "lr": 5e-05, "grad_norm": 0.2496311793376771, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111402662411, "tokens_per_second": 120317.83418315221, "data_wait_s": 1.1202995665371418e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 761200, "loss": 0.5246676206588745, "lr": 5e-05, "grad_norm": 0.22605731974150353, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111417437559, "tokens_per_second": 153414.49861115217, "data_wait_s": 1.658000110182911e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 761300, "loss": 0.5713633894920349, "lr": 5e-05, "grad_norm": 0.23163489212137198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111432132086, "tokens_per_second": 151258.01560694363, "data_wait_s": 1.29489999380894e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:27d476213436fec8244aa0d7864a2832e55ae73f6fa9867b37223e6ee5476507
-size 879045
+oid sha256:3b28b116aff10edbfba6c5b826db0748c0aa99b71bfc74e2b9cabfb41b25bac7
+size 881163