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": 76318.94226167015,
-      "required_pages_per_second_upper_estimate": 0.005279921510005849
+      "required_token_positions_per_second": 77203.97949752082,
+      "required_pages_per_second_upper_estimate": 0.005341150439551312
     },
     {
       "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": 9539.867782708769,
-      "required_pages_per_second_upper_estimate": 0.003994143425527355
+      "required_token_positions_per_second": 9650.497437190103,
+      "required_pages_per_second_upper_estimate": 0.004040461751648801
     },
     {
       "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": 19079.735565417537,
-      "required_pages_per_second_upper_estimate": 0.003973805751813134
+      "required_token_positions_per_second": 19300.994874380205,
+      "required_pages_per_second_upper_estimate": 0.004019888230869693
     },
     {
       "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": 28619.603348126304,
-      "required_pages_per_second_upper_estimate": 0.0017157015211161485
+      "required_token_positions_per_second": 28951.49231157031,
+      "required_pages_per_second_upper_estimate": 0.001735597757709512
     },
     {
       "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": 28619.603348126304,
-      "required_pages_per_second_upper_estimate": 0.002225048275985558
+      "required_token_positions_per_second": 28951.49231157031,
+      "required_pages_per_second_upper_estimate": 0.0022508511830680584
     },
     {
       "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": 28619.603348126304,
-      "required_pages_per_second_upper_estimate": 0.36710154241385184
+      "required_token_positions_per_second": 28951.49231157031,
+      "required_pages_per_second_upper_estimate": 0.37135865768230664
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.07350226860933162,
-    "producer_p95_seconds": 0.20136206700044568,
-    "gpu_update_seconds": 0.9660511245977867,
-    "producer_batches_per_second": 13.605022251966833,
-    "gpu_batches_per_second": 1.0351419035057257,
-    "producer_headroom": 13.143147046690473,
+    "producer_batch_seconds": 0.2611032896640211,
+    "producer_p95_seconds": 0.20631458099887823,
+    "gpu_update_seconds": 0.954976679697807,
+    "producer_batches_per_second": 3.8299019567572907,
+    "gpu_batches_per_second": 1.04714598927844,
+    "producer_headroom": 3.657467054232211,
     "sustainable": true,
-    "buffer_seconds": 60.86122084966056,
+    "buffer_seconds": 60.16353082096184,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 436602.4299373324,
-    "upload_bytes_per_second": 5690849.82321252,
+    "generation_bytes_per_second": 402983.92484912527,
+    "upload_bytes_per_second": 5738374.5109207425,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.68471788015346,
+    "estimated_full_shards_per_hour": 86.47096928696936,
     "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": 1.4037659219993657,
-      "bytes": 10492690,
-      "pages": 42,
-      "shards": 3
-    },
     {
       "seconds": 0.9956328799999028,
       "bytes": 1276481,
@@ -562,6 +556,12 @@
       "bytes": 857193,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.0425618379995285,
+      "bytes": 11991809,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -145,3 +145,4 @@
 {"step": 539500, "loss": 0.5756470561027527, "lr": 0.0003, "grad_norm": 0.16138214906417178, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78964650115, "tokens_per_second": 151584.61324195223, "data_wait_s": 1.0184001439483836e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 539600, "loss": 0.5874589085578918, "lr": 0.0003, "grad_norm": 0.16471172094975167, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78979216356, "tokens_per_second": 152348.29667662186, "data_wait_s": 6.40399957774207e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 539700, "loss": 0.5510649085044861, "lr": 0.0003, "grad_norm": 0.16384735945024742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78993891745, "tokens_per_second": 153414.08955337649, "data_wait_s": 1.0887999451369978e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 539800, "loss": 0.6331631541252136, "lr": 0.0003, "grad_norm": 0.18237055770298055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79008423707, "tokens_per_second": 152120.11196001043, "data_wait_s": 8.068000170169398e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:41a9239301f237da7405f73edabfca8dcd3a227ad69716ab4061e5d49b630cda
-size 313077
+oid sha256:0a92c834ac5c2d44cc527fbfd61f6876477d59098db5a37904bb37d727b85a24
+size 315195