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": 76620.07702799016,
-      "required_pages_per_second_upper_estimate": 0.005300754711868792
+      "required_token_positions_per_second": 76509.39123526706,
+      "required_pages_per_second_upper_estimate": 0.005293097211901784
     },
     {
       "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": 9577.50962849877,
-      "required_pages_per_second_upper_estimate": 0.004009903280308504
+      "required_token_positions_per_second": 9563.673904408382,
+      "required_pages_per_second_upper_estimate": 0.004004110551554632
     },
     {
       "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": 19155.01925699754,
-      "required_pages_per_second_upper_estimate": 0.003989485359404792
+      "required_token_positions_per_second": 19127.347808816765,
+      "required_pages_per_second_upper_estimate": 0.003983722126493898
     },
     {
       "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": 28732.52888549631,
-      "required_pages_per_second_upper_estimate": 0.0017224712346541687
+      "required_token_positions_per_second": 28691.02171322515,
+      "required_pages_per_second_upper_estimate": 0.001719982943054295
     },
     {
       "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": 28732.52888549631,
-      "required_pages_per_second_upper_estimate": 0.002233827739809131
+      "required_token_positions_per_second": 28691.02171322515,
+      "required_pages_per_second_upper_estimate": 0.002230600739735804
     },
     {
       "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": 28732.52888549631,
-      "required_pages_per_second_upper_estimate": 0.368550029957239
+      "required_token_positions_per_second": 28691.02171322515,
+      "required_pages_per_second_upper_estimate": 0.36801762051827386
     }
   ],
   "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": 0.09847208613263092,
-    "producer_p95_seconds": 0.27197384500323096,
-    "gpu_update_seconds": 0.9622543184479747,
-    "producer_batches_per_second": 10.15516213044488,
-    "gpu_batches_per_second": 1.0392263051756478,
-    "producer_headroom": 9.771848614559921,
+    "producer_batch_seconds": 0.11352637590624681,
+    "producer_p95_seconds": 0.2615216939993843,
+    "gpu_update_seconds": 0.9636464074493253,
+    "producer_batches_per_second": 8.808525701779006,
+    "gpu_batches_per_second": 1.0377250330304235,
+    "producer_headroom": 8.488304147444387,
     "sustainable": true,
-    "buffer_seconds": 60.622022062222406,
+    "buffer_seconds": 60.70972366930749,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 356711.13277459226,
-    "upload_bytes_per_second": 3848587.7147090696,
+    "generation_bytes_per_second": 404413.5880052211,
+    "upload_bytes_per_second": 3827339.350250768,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.54190528324438,
+    "estimated_full_shards_per_hour": 86.77774171941256,
     "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": 1.1709360159984499,
-      "bytes": 6311701,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.1361314889982168,
       "bytes": 6780795,
@@ -562,6 +556,12 @@
       "bytes": 5804640,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.8439256270030455,
+      "bytes": 2654364,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -270,3 +270,4 @@
 {"step": 552000, "loss": 0.611603856086731, "lr": 0.0003, "grad_norm": 0.1777720764578613, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80795224140, "tokens_per_second": 152007.90497875676, "data_wait_s": 2.2330998035613447e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 552100, "loss": 0.6218159198760986, "lr": 0.0003, "grad_norm": 0.1657206908340373, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80809896889, "tokens_per_second": 152722.40076258013, "data_wait_s": 1.763199907145463e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 552200, "loss": 0.5662043690681458, "lr": 0.0003, "grad_norm": 0.16982648571386338, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80824575209, "tokens_per_second": 153258.87326916007, "data_wait_s": 1.0771000233944505e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 552300, "loss": 0.5912648439407349, "lr": 0.0003, "grad_norm": 0.14267824863647247, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80839119291, "tokens_per_second": 150291.8241835844, "data_wait_s": 1.692500154604204e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ac552c9fcdb243dd5465576dc7f5ba2d0679ee73df8a55a762d71323d2340253
-size 580176
+oid sha256:1cff0bab8a5318a23e4477b9718be8845598bc0427426ce9d4d83d6b088ab77a
+size 582294