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": 77242.76469406058,
-      "required_pages_per_second_upper_estimate": 0.005343833689441988
+      "required_token_positions_per_second": 76765.28199226498,
+      "required_pages_per_second_upper_estimate": 0.005310800328219259
     },
     {
       "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": 9655.345586757572,
-      "required_pages_per_second_upper_estimate": 0.004042491570631825
+      "required_token_positions_per_second": 9595.660249033122,
+      "required_pages_per_second_upper_estimate": 0.00401750256609818
     },
     {
       "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": 19310.691173515144,
-      "required_pages_per_second_upper_estimate": 0.0040219077142708975
+      "required_token_positions_per_second": 19191.320498066245,
+      "required_pages_per_second_upper_estimate": 0.003997045950591298
     },
     {
       "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": 28966.036760272713,
-      "required_pages_per_second_upper_estimate": 0.001736469675201134
+      "required_token_positions_per_second": 28786.98074709937,
+      "required_pages_per_second_upper_estimate": 0.0017257355406140936
     },
     {
       "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": 28966.036760272713,
-      "required_pages_per_second_upper_estimate": 0.0022519819499804013
+      "required_token_positions_per_second": 28786.98074709937,
+      "required_pages_per_second_upper_estimate": 0.0022380611325403415
     },
     {
       "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": 28966.036760272713,
-      "required_pages_per_second_upper_estimate": 0.3715452182536488
+      "required_token_positions_per_second": 28786.98074709937,
+      "required_pages_per_second_upper_estimate": 0.3692484799720292
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.13130447249216814,
-    "producer_p95_seconds": 0.132896368000047,
-    "gpu_update_seconds": 0.9544971660713896,
-    "producer_batches_per_second": 7.615886808879618,
-    "gpu_batches_per_second": 1.0476720471742156,
-    "producer_headroom": 7.269342376196073,
+    "producer_batch_seconds": 0.0866183350937515,
+    "producer_p95_seconds": 0.14069177400006083,
+    "gpu_update_seconds": 0.9604341713670638,
+    "producer_batches_per_second": 11.544899805770319,
+    "gpu_batches_per_second": 1.041195773549601,
+    "producer_headroom": 11.08811627847079,
     "sustainable": true,
-    "buffer_seconds": 60.13332146249754,
+    "buffer_seconds": 60.50735279612502,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 49,
+    "upload_pages_per_commit": 52,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 314464.7946407608,
-    "upload_bytes_per_second": 7119838.715995101,
+    "generation_bytes_per_second": 382708.29017527483,
+    "upload_bytes_per_second": 5944382.736932924,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.47682456414336,
+    "estimated_full_shards_per_hour": 82.12029007857974,
     "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": [
@@ -190,6 +190,12 @@
       "bytes": 16191062,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 0.8292768119999891,
+      "bytes": 2252671,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -3,3 +3,4 @@
 {"step": 440500, "loss": 0.564326822757721, "lr": 0.0003, "grad_norm": 0.1553253692505358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64476119199, "tokens_per_second": 151680.3247810244, "data_wait_s": 1.7253999999411462e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 440600, "loss": 0.5960808992385864, "lr": 0.0003, "grad_norm": 0.18775933079072962, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64490815565, "tokens_per_second": 143499.93440440673, "data_wait_s": 1.7190000107802916e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 440700, "loss": 0.5901668667793274, "lr": 0.0003, "grad_norm": 0.1588032309510571, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64505460141, "tokens_per_second": 151138.92304691722, "data_wait_s": 1.7090000028474606e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 440800, "loss": 0.7072717547416687, "lr": 0.0003, "grad_norm": 0.17982194413959549, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64520273027, "tokens_per_second": 153151.86996610765, "data_wait_s": 1.6521000134162023e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3f1a31cd5e2a177eec18be9e845cdac248f67e974e97650785ccd7ffe7851b7d
-size 10755
+oid sha256:5338fb72090707ed3535fad834056ae9e9ae098b2d935de1854477855b1671b7
+size 12873