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": 76910.4344428631,
-      "required_pages_per_second_upper_estimate": 0.005320842311551725
+      "required_token_positions_per_second": 76750.9427922613,
+      "required_pages_per_second_upper_estimate": 0.0053098083090914764
     },
     {
       "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": 9613.804305357888,
-      "required_pages_per_second_upper_estimate": 0.0040250991035903775
+      "required_token_positions_per_second": 9593.867849032662,
+      "required_pages_per_second_upper_estimate": 0.004016752125647555
     },
     {
       "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": 19227.608610715775,
-      "required_pages_per_second_upper_estimate": 0.00400460380747431
+      "required_token_positions_per_second": 19187.735698065324,
+      "required_pages_per_second_upper_estimate": 0.00399629933128864
     },
     {
       "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": 28841.41291607366,
-      "required_pages_per_second_upper_estimate": 0.0017289986660309925
+      "required_token_positions_per_second": 28781.603547097984,
+      "required_pages_per_second_upper_estimate": 0.0017254131856845195
     },
     {
       "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": 28841.41291607366,
-      "required_pages_per_second_upper_estimate": 0.002242292994256284
+      "required_token_positions_per_second": 28781.603547097984,
+      "required_pages_per_second_upper_estimate": 0.0022376430788920367
     },
     {
       "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": 28841.41291607366,
-      "required_pages_per_second_upper_estimate": 0.36994667739092185
+      "required_token_positions_per_second": 28781.603547097984,
+      "required_pages_per_second_upper_estimate": 0.369179507023999
     }
   ],
   "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": 0.09407934603905233,
-    "producer_p95_seconds": 0.2824194999993779,
-    "gpu_update_seconds": 0.9586215516019828,
-    "producer_batches_per_second": 10.629325586350273,
-    "gpu_batches_per_second": 1.043164529661229,
-    "producer_headroom": 10.189500586069753,
+    "producer_batch_seconds": 0.20074863879705163,
+    "producer_p95_seconds": 0.29929958100183285,
+    "gpu_update_seconds": 0.9606136070478851,
+    "producer_batches_per_second": 4.98135382631888,
+    "gpu_batches_per_second": 1.0410012857023287,
+    "producer_headroom": 4.785156267081964,
     "sustainable": true,
-    "buffer_seconds": 60.393157750924914,
+    "buffer_seconds": 60.518657244016765,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 503933.5921704239,
-    "upload_bytes_per_second": 5771119.549278284,
+    "generation_bytes_per_second": 481386.61246356735,
+    "upload_bytes_per_second": 5806526.815700199,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 108.13241790613687,
+    "estimated_full_shards_per_hour": 103.29436092787041,
     "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": 1.0631770070031052,
-      "bytes": 6064387,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.2800072050013114,
       "bytes": 12000660,
@@ -562,6 +556,12 @@
       "bytes": 6872118,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.3191045509993273,
+      "bytes": 10179142,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -298,3 +298,4 @@
 {"step": 586900, "loss": 0.5674201250076294, "lr": 0.0003, "grad_norm": 0.151298944953679, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85904721763, "tokens_per_second": 152233.33595929542, "data_wait_s": 6.4159976318478584e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 587000, "loss": 0.5778745412826538, "lr": 0.0003, "grad_norm": 0.16915777713992394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85919437460, "tokens_per_second": 152308.90628709272, "data_wait_s": 1.4666999049950391e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 587100, "loss": 0.5684595704078674, "lr": 0.0003, "grad_norm": 0.17902361592404403, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85934066337, "tokens_per_second": 152722.94810485, "data_wait_s": 1.0110998118761927e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 587200, "loss": 0.5829356908798218, "lr": 0.0003, "grad_norm": 0.17970318356402584, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85948598787, "tokens_per_second": 150609.5345376927, "data_wait_s": 1.430700285709463e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:49764ada3ad064320414f24d274f7ea66fd2f6eda33eb91f96b1c476b6b3d155
-size 640263
+oid sha256:92051790beb078ed7368c510b5b38a2953caa7dab6557efa301ed2bbc45f97f2
+size 642381