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": 76952.91239146466,
-      "required_pages_per_second_upper_estimate": 0.005323781034598401
+      "required_token_positions_per_second": 77164.09611100632,
+      "required_pages_per_second_upper_estimate": 0.005338391214330033
     },
     {
       "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": 9619.114048933083,
-      "required_pages_per_second_upper_estimate": 0.004027322182345258
+      "required_token_positions_per_second": 9645.51201387579,
+      "required_pages_per_second_upper_estimate": 0.004038374459014576
     },
     {
       "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": 19238.228097866166,
-      "required_pages_per_second_upper_estimate": 0.004006815566593028
+      "required_token_positions_per_second": 19291.02402775158,
+      "required_pages_per_second_upper_estimate": 0.004017811566465859
     },
     {
       "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": 28857.34214679925,
-      "required_pages_per_second_upper_estimate": 0.0017299535990904647
+      "required_token_positions_per_second": 28936.53604162737,
+      "required_pages_per_second_upper_estimate": 0.0017347011521633346
     },
     {
       "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": 28857.34214679925,
-      "required_pages_per_second_upper_estimate": 0.0022435314218799173
+      "required_token_positions_per_second": 28936.53604162737,
+      "required_pages_per_second_upper_estimate": 0.0022496883988656737
     },
     {
       "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": 28857.34214679925,
-      "required_pages_per_second_upper_estimate": 0.3701510004591943
+      "required_token_positions_per_second": 28936.53604162737,
+      "required_pages_per_second_upper_estimate": 0.37116681471027013
     }
   ],
   "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": 0.0949183044297186,
-    "producer_p95_seconds": 0.1695962559997497,
-    "gpu_update_seconds": 0.9580923932409561,
-    "producer_batches_per_second": 10.535375721344042,
-    "gpu_batches_per_second": 1.0437406737123571,
-    "producer_headroom": 10.093863338555177,
+    "producer_batch_seconds": 0.21388095956251618,
+    "producer_p95_seconds": 0.17199944499952835,
+    "gpu_update_seconds": 0.9554702732983066,
+    "producer_batches_per_second": 4.675498006206138,
+    "gpu_batches_per_second": 1.0466050362278416,
+    "producer_headroom": 4.467299357795466,
     "sustainable": true,
-    "buffer_seconds": 60.359820774180236,
+    "buffer_seconds": 60.19462721779332,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 451828.2699052624,
-    "upload_bytes_per_second": 2224958.1321611064,
+    "generation_bytes_per_second": 411163.47786046617,
+    "upload_bytes_per_second": 2286669.0387693066,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.95182869785694,
+    "estimated_full_shards_per_hour": 88.22611095295419,
     "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": [
@@ -316,6 +316,12 @@
       "bytes": 1194300,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.6349133659996369,
+      "bytes": 7762216,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -60,3 +60,4 @@
 {"step": 651000, "loss": 0.5534883737564087, "lr": 0.0001, "grad_norm": 0.20804037514537232, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95289164028, "tokens_per_second": 153557.97579697263, "data_wait_s": 1.0865999684028793e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 651100, "loss": 0.5618123412132263, "lr": 0.0001, "grad_norm": 0.17966431658741164, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95303841629, "tokens_per_second": 152674.4445997139, "data_wait_s": 9.499000043433625e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 651200, "loss": 0.5152514576911926, "lr": 0.0001, "grad_norm": 0.19390245787186547, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95318582360, "tokens_per_second": 153970.00772675822, "data_wait_s": 1.3885000043956097e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 651300, "loss": 0.5093046426773071, "lr": 0.0001, "grad_norm": 0.1805517497239884, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95333195261, "tokens_per_second": 152386.52056084806, "data_wait_s": 1.4483000086329412e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3e978af5099df46f17341afa7d16d8eabd2722176a9f451f8cc06c9556f180a2
-size 132264
+oid sha256:55383b9d273a31e18eb351f1274ac35f8635e3537c2a8b6dd6af889dbf5d967e
+size 134382