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": 77097.26337254171,
-      "required_pages_per_second_upper_estimate": 0.005333767570409733
+      "required_token_positions_per_second": 76647.68028587635,
+      "required_pages_per_second_upper_estimate": 0.005302664369297741
     },
     {
       "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": 9637.157921567714,
-      "required_pages_per_second_upper_estimate": 0.004034876775018469
+      "required_token_positions_per_second": 9580.960035734544,
+      "required_pages_per_second_upper_estimate": 0.0040113478937654256
     },
     {
       "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": 19274.315843135428,
-      "required_pages_per_second_upper_estimate": 0.004014331692234773
+      "required_token_positions_per_second": 19161.92007146909,
+      "required_pages_per_second_upper_estimate": 0.003990922617072499
     },
     {
       "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": 28911.473764703143,
-      "required_pages_per_second_upper_estimate": 0.0017331987069296084
+      "required_token_positions_per_second": 28742.880107203633,
+      "required_pages_per_second_upper_estimate": 0.0017230917745901222
     },
     {
       "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": 28911.473764703143,
-      "required_pages_per_second_upper_estimate": 0.0022477399170721213
+      "required_token_positions_per_second": 28742.880107203633,
+      "required_pages_per_second_upper_estimate": 0.0022346325017666624
     },
     {
       "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": 28911.473764703143,
-      "required_pages_per_second_upper_estimate": 0.37084534273166253
+      "required_token_positions_per_second": 28742.880107203633,
+      "required_pages_per_second_upper_estimate": 0.3686828043150246
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.3281743131250181,
-    "producer_p95_seconds": 0.14285689700000148,
-    "gpu_update_seconds": 0.9562985347967399,
-    "producer_batches_per_second": 3.047161097032752,
-    "gpu_batches_per_second": 1.0456985591978856,
-    "producer_headroom": 2.913995692382047,
+    "producer_batch_seconds": 0.06792304984375264,
+    "producer_p95_seconds": 0.15757650899968212,
+    "gpu_update_seconds": 0.9619077801834747,
+    "producer_batches_per_second": 14.722542675871571,
+    "gpu_batches_per_second": 1.0396006983218906,
+    "producer_headroom": 14.161728344004096,
     "sustainable": true,
-    "buffer_seconds": 60.246807692194615,
+    "buffer_seconds": 60.600190151558905,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 362577.40623319143,
-    "upload_bytes_per_second": 5326621.676999357,
+    "generation_bytes_per_second": 362975.0652812295,
+    "upload_bytes_per_second": 5310956.942838345,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.80067100760276,
+    "estimated_full_shards_per_hour": 77.88599938228286,
     "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": [
@@ -292,6 +292,12 @@
       "bytes": 10812914,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.5107661739998548,
+      "bytes": 7696973,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -20,3 +20,4 @@
 {"step": 559100, "loss": 0.5916821956634521, "lr": 0.0003, "grad_norm": 0.20054137954538745, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81834664568, "tokens_per_second": 153743.3016488649, "data_wait_s": 1.3612999737233622e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 559200, "loss": 0.5428046584129333, "lr": 0.0003, "grad_norm": 0.1846452623330113, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81849262503, "tokens_per_second": 151677.33703198694, "data_wait_s": 1.6266000329778763e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 559300, "loss": 0.5671708583831787, "lr": 0.0003, "grad_norm": 0.1635655099334, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81863888299, "tokens_per_second": 151940.756692194, "data_wait_s": 1.510600031906506e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 559400, "loss": 0.5463201999664307, "lr": 0.0003, "grad_norm": 0.16494876638535577, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81878508084, "tokens_per_second": 151462.2509951918, "data_wait_s": 1.4911000107531436e-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:e8eb31dd5f1fdabb7f9d01f190a474375661d8af32783f7eccb07d7287a28490
-size 46761
+oid sha256:421424900198a07af10542585e0355558cc8068529d973dde2d9105b30276909
+size 48879