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": 76818.81403360202,
-      "required_pages_per_second_upper_estimate": 0.005314503799050405
+      "required_token_positions_per_second": 76510.86186016112,
+      "required_pages_per_second_upper_estimate": 0.005293198953144276
     },
     {
       "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": 9602.351754200252,
-      "required_pages_per_second_upper_estimate": 0.004020304159577139
+      "required_token_positions_per_second": 9563.85773252014,
+      "required_pages_per_second_upper_estimate": 0.004004187516546257
     },
     {
       "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": 19204.703508400504,
-      "required_pages_per_second_upper_estimate": 0.003999833278710208
+      "required_token_positions_per_second": 19127.71546504028,
+      "required_pages_per_second_upper_estimate": 0.00398379869958951
     },
     {
       "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": 28807.05526260076,
-      "required_pages_per_second_upper_estimate": 0.0017269389771663915
+      "required_token_positions_per_second": 28691.57319756042,
+      "required_pages_per_second_upper_estimate": 0.0017200160036981267
     },
     {
       "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": 28807.05526260076,
-      "required_pages_per_second_upper_estimate": 0.0022396218378221123
+      "required_token_positions_per_second": 28691.57319756042,
+      "required_pages_per_second_upper_estimate": 0.002230643615217148
     },
     {
       "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": 28807.05526260076,
-      "required_pages_per_second_upper_estimate": 0.3695059743025456
+      "required_token_positions_per_second": 28691.57319756042,
+      "required_pages_per_second_upper_estimate": 0.368024694367189
     }
   ],
   "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": 0.0836552607656813,
-    "producer_p95_seconds": 0.2295517380007368,
-    "gpu_update_seconds": 0.9597648821778212,
-    "producer_batches_per_second": 11.953820845780443,
-    "gpu_batches_per_second": 1.0419218483290205,
-    "producer_headroom": 11.472857455625249,
+    "producer_batch_seconds": 0.11482742489818065,
+    "producer_p95_seconds": 0.24376358700101264,
+    "gpu_update_seconds": 0.9636278850805868,
+    "producer_batches_per_second": 8.708720942638191,
+    "gpu_batches_per_second": 1.0377449796571334,
+    "producer_headroom": 8.391966343711456,
     "sustainable": true,
-    "buffer_seconds": 60.465187577202734,
+    "buffer_seconds": 60.70855676007697,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 262070.08536985546,
-    "upload_bytes_per_second": 5173550.580958508,
+    "generation_bytes_per_second": 274257.6599293496,
+    "upload_bytes_per_second": 5117282.01190856,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.23413964101551,
+    "estimated_full_shards_per_hour": 58.849309429267564,
     "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": 1.163782216999607,
-      "bytes": 7671664,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.1290235629985546,
       "bytes": 1186173,
@@ -562,6 +556,12 @@
       "bytes": 9839539,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.0725541200008593,
+      "bytes": 2601868,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -198,3 +198,4 @@
 {"step": 576900, "loss": 0.63382887840271, "lr": 0.0003, "grad_norm": 0.1661946634264805, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84441024108, "tokens_per_second": 152747.27800623022, "data_wait_s": 1.2516997230704874e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 577000, "loss": 0.5794296860694885, "lr": 0.0003, "grad_norm": 0.16336871154826024, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84455571549, "tokens_per_second": 151905.4603598286, "data_wait_s": 1.8733000615611672e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 577100, "loss": 0.5727930665016174, "lr": 0.0003, "grad_norm": 0.1746464669962155, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84470083327, "tokens_per_second": 150917.30534864785, "data_wait_s": 1.0232000931864604e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 577200, "loss": 0.6069290041923523, "lr": 0.0003, "grad_norm": 0.17648931833721537, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84484589701, "tokens_per_second": 150581.32165991655, "data_wait_s": 1.3238000974524766e-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:ecfbeb7850b5142bec273d9acc02e27a1384062fd7bd4371dda43127331736a8
-size 426897
+oid sha256:e528e38998ea08eafc7a8f4ce46c81de206a47a843117fec174f862789796f57
+size 429015