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": 77064.64306717341,
-      "required_pages_per_second_upper_estimate": 0.005331510821994818
+      "required_token_positions_per_second": 76738.7231906564,
+      "required_pages_per_second_upper_estimate": 0.005308962928699068
     },
     {
       "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": 9633.080383396677,
-      "required_pages_per_second_upper_estimate": 0.004033169594934936
+      "required_token_positions_per_second": 9592.34039883205,
+      "required_pages_per_second_upper_estimate": 0.004016112614145346
     },
     {
       "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": 19266.160766793353,
-      "required_pages_per_second_upper_estimate": 0.004012633204896545
+      "required_token_positions_per_second": 19184.6807976641,
+      "required_pages_per_second_upper_estimate": 0.003995663076098206
     },
     {
       "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": 28899.24115019003,
-      "required_pages_per_second_upper_estimate": 0.0017324653803676682
+      "required_token_positions_per_second": 28777.02119649615,
+      "required_pages_per_second_upper_estimate": 0.0017251384807627837
     },
     {
       "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": 28899.24115019003,
-      "required_pages_per_second_upper_estimate": 0.0022467888850993115
+      "required_token_positions_per_second": 28777.02119649615,
+      "required_pages_per_second_upper_estimate": 0.0022372868212883743
     },
     {
       "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": 28899.24115019003,
-      "required_pages_per_second_upper_estimate": 0.3706884358870465
+      "required_token_positions_per_second": 28777.02119649615,
+      "required_pages_per_second_upper_estimate": 0.36912072955062336
     }
   ],
   "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": 0.1021241071328518,
-    "producer_p95_seconds": 0.31390461699993466,
-    "gpu_update_seconds": 0.9567033215963249,
-    "producer_batches_per_second": 9.792007275021893,
-    "gpu_batches_per_second": 1.0452561179900908,
-    "producer_headroom": 9.368045885108822,
+    "producer_batch_seconds": 0.2100383155311647,
+    "producer_p95_seconds": 0.2853320079993864,
+    "gpu_update_seconds": 0.9607665717453195,
+    "producer_batches_per_second": 4.761036087492445,
+    "gpu_batches_per_second": 1.0408355467482693,
+    "producer_headroom": 4.5742443197358655,
     "sustainable": true,
-    "buffer_seconds": 60.27230926056847,
+    "buffer_seconds": 60.52829401995513,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 330577.3062294218,
-    "upload_bytes_per_second": 5769569.977629003,
+    "generation_bytes_per_second": 346614.60156898265,
+    "upload_bytes_per_second": 5746102.557139049,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.93419447099677,
+    "estimated_full_shards_per_hour": 74.37542472173796,
     "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": 0.9885955910031043,
-      "bytes": 2380823,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.9730731119998381,
       "bytes": 6090581,
@@ -562,6 +556,12 @@
       "bytes": 11349081,
       "pages": 37,
       "shards": 4
+    },
+    {
+      "seconds": 1.0920391100007691,
+      "bytes": 1249324,
+      "pages": 36,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -281,3 +281,4 @@
 {"step": 585200, "loss": 0.5328781008720398, "lr": 0.0003, "grad_norm": 0.14028533368816135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85656318075, "tokens_per_second": 150663.712994855, "data_wait_s": 1.3040000339969993e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 585300, "loss": 0.5749945044517517, "lr": 0.0003, "grad_norm": 0.1715119345751613, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85670779550, "tokens_per_second": 150466.36007464418, "data_wait_s": 1.2612003047252074e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 585400, "loss": 0.6363208293914795, "lr": 0.0003, "grad_norm": 0.16628964172583055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85685347083, "tokens_per_second": 151389.35578837435, "data_wait_s": 1.2720000086119398e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 585500, "loss": 0.6167413592338562, "lr": 0.0003, "grad_norm": 0.15069580673872307, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85699987861, "tokens_per_second": 152128.00421720796, "data_wait_s": 1.2340999091975391e-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:a54cc1778342100c5d3d7968ae4611f77bcb2916ecef00b76d240ee23fe01a3b
-size 604257
+oid sha256:45bd9c59ef67ed046871ec69f6ffa1cebd3c4652ee943018ad1ce3fce8975757
+size 606375