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": 77128.24863329732,
-      "required_pages_per_second_upper_estimate": 0.005335911202644516
+      "required_token_positions_per_second": 77319.56427805276,
+      "required_pages_per_second_upper_estimate": 0.005349146862862152
     },
     {
       "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": 9641.031079162165,
-      "required_pages_per_second_upper_estimate": 0.004036498385222538
+      "required_token_positions_per_second": 9664.945534756594,
+      "required_pages_per_second_upper_estimate": 0.0040465108683893055
     },
     {
       "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": 19282.06215832433,
-      "required_pages_per_second_upper_estimate": 0.0040159450454045524
+      "required_token_positions_per_second": 19329.89106951319,
+      "required_pages_per_second_upper_estimate": 0.0040259065462719835
     },
     {
       "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": 28923.093237486497,
-      "required_pages_per_second_upper_estimate": 0.0017338952765810628
+      "required_token_positions_per_second": 28994.83660426978,
+      "required_pages_per_second_upper_estimate": 0.0017381961818739426
     },
     {
       "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": 28923.093237486497,
-      "required_pages_per_second_upper_estimate": 0.0022486432799723676
+      "required_token_positions_per_second": 28994.83660426978,
+      "required_pages_per_second_upper_estimate": 0.002254221011174047
     },
     {
       "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": 28923.093237486497,
-      "required_pages_per_second_upper_estimate": 0.3709943848525096
+      "required_token_positions_per_second": 28994.83660426978,
+      "required_pages_per_second_upper_estimate": 0.37191463172957995
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.16861545439826386,
-    "producer_p95_seconds": 0.23194225399856805,
-    "gpu_update_seconds": 0.955914354422027,
-    "producer_batches_per_second": 5.930654479855889,
-    "gpu_batches_per_second": 1.0461188236938115,
-    "producer_headroom": 5.669197748411545,
+    "producer_batch_seconds": 0.11988631587507825,
+    "producer_p95_seconds": 0.2385293189981894,
+    "gpu_update_seconds": 0.9535490879754968,
+    "producer_batches_per_second": 8.341235550536075,
+    "gpu_batches_per_second": 1.0487137081984153,
+    "producer_headroom": 7.953777551802464,
     "sustainable": true,
-    "buffer_seconds": 60.2226043285877,
+    "buffer_seconds": 60.073592542456296,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 342824.15874215215,
+    "generation_bytes_per_second": 343421.98595938255,
     "upload_bytes_per_second": 4791392.151073171,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.56208392809319,
+    "estimated_full_shards_per_hour": 73.69036373220545,
     "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": [

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -218,3 +218,4 @@
 {"step": 946800, "loss": 0.49404534697532654, "lr": 3e-05, "grad_norm": 0.35579109683225085, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138585287594, "tokens_per_second": 152540.61911892937, "data_wait_s": 1.336399873252958e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 946900, "loss": 0.5574998259544373, "lr": 3e-05, "grad_norm": 0.33907311921784394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138600019929, "tokens_per_second": 153279.65545405858, "data_wait_s": 1.207899913424626e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 947000, "loss": 0.5088230967521667, "lr": 3e-05, "grad_norm": 0.32660654293101216, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138614775562, "tokens_per_second": 154054.09789012928, "data_wait_s": 1.0375002602813765e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 947100, "loss": 0.47206395864486694, "lr": 3e-05, "grad_norm": 0.33087760256377796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138629409034, "tokens_per_second": 152035.63705868647, "data_wait_s": 1.5099998563528061e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:abbdd073b4f1f46e148466023cfd605dab3e6eb93b00d1501068ea42ddded74a
-size 469257
+oid sha256:f3d398aea2059618e90e65c7d1163c7814f2ddc3b27a3a08077e262d9870695a
+size 471375