Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75467.55172386288,
-      "required_pages_per_second_upper_estimate": 0.006541336112663273
+      "required_token_positions_per_second": 76095.36622040457,
+      "required_pages_per_second_upper_estimate": 0.006595753482042212
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9433.44396548286,
-      "required_pages_per_second_upper_estimate": 0.003763141840387291
+      "required_token_positions_per_second": 9511.920777550571,
+      "required_pages_per_second_upper_estimate": 0.0037944474140540014
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18866.88793096572,
-      "required_pages_per_second_upper_estimate": 0.00436452247694211
+      "required_token_positions_per_second": 19023.841555101142,
+      "required_pages_per_second_upper_estimate": 0.004400830935596399
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28300.33189644858,
-      "required_pages_per_second_upper_estimate": 0.0017233417321274253
+      "required_token_positions_per_second": 28535.762332651713,
+      "required_pages_per_second_upper_estimate": 0.001737678210484159
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28300.33189644858,
-      "required_pages_per_second_upper_estimate": 0.0021786378897213055
+      "required_token_positions_per_second": 28535.762332651713,
+      "required_pages_per_second_upper_estimate": 0.0021967619764133813
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28300.33189644858,
-      "required_pages_per_second_upper_estimate": 0.3513954070358789
+      "required_token_positions_per_second": 28535.762332651713,
+      "required_pages_per_second_upper_estimate": 0.3543186651185382
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.153689610234494,
-    "producer_p95_seconds": 0.4033702369997627,
-    "gpu_update_seconds": 0.9769496732817313,
-    "producer_batches_per_second": 6.5066207043809685,
-    "gpu_batches_per_second": 1.0235941802824284,
-    "producer_headroom": 6.356640971313136,
+    "producer_batch_seconds": 0.10179032871093341,
+    "producer_p95_seconds": 0.4072658369987039,
+    "gpu_update_seconds": 0.9688894825271269,
+    "producer_batches_per_second": 9.824116030117397,
+    "gpu_batches_per_second": 1.0321094593696367,
+    "producer_headroom": 9.518482696706897,
     "sustainable": true,
-    "buffer_seconds": 60.570879743467344,
+    "buffer_seconds": 60.071147916681866,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326625.2953772913,
-    "upload_bytes_per_second": 6236135.973197361,
+    "generation_bytes_per_second": 326417.2543513872,
+    "upload_bytes_per_second": 6182411.037728616,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.08618494023374,
+    "estimated_full_shards_per_hour": 70.04154417902195,
     "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": 1.2022184809975442,
-      "bytes": 7326926,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 0.7280326340005558,
       "bytes": 1168675,
@@ -562,6 +556,12 @@
       "bytes": 7423511,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.7559060139974463,
+      "bytes": 900379,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -378,3 +378,4 @@
 {"step": 77800, "loss": 0.6180443204939365, "lr": 0.0003, "grad_norm": 0.14828734039842262, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11385783307, "tokens_per_second": 149977.69468966572, "data_wait_s": 1.0100004146806896e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 77900, "loss": 0.648516247048974, "lr": 0.0003, "grad_norm": 0.133658089222264, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11400499219, "tokens_per_second": 149770.91388315082, "data_wait_s": 9.019000572152436e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 78000, "loss": 0.6717341057956219, "lr": 0.0003, "grad_norm": 0.1393901915487311, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11415141762, "tokens_per_second": 149919.69667152414, "data_wait_s": 1.1878997611347586e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 78100, "loss": 0.6047646552324295, "lr": 0.0003, "grad_norm": 0.13520182350403326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11429812036, "tokens_per_second": 148336.8116493828, "data_wait_s": 1.0003001079894602e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9f9af6adfa352fc1346741d9725f72d104cacdbc0878ca178e248b70edc9584d
-size 810486
+oid sha256:cf1d9f57533383d5eda987a837ebc6f977f1e22b1aa422e50b68ef7f9a7a2350
+size 812604