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": 75459.7910829304,
-      "required_pages_per_second_upper_estimate": 0.005220483437955247
+      "required_token_positions_per_second": 75745.46080173945,
+      "required_pages_per_second_upper_estimate": 0.005240246731947527
     },
     {
       "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": 9432.4738853663,
-      "required_pages_per_second_upper_estimate": 0.003949179843349659
+      "required_token_positions_per_second": 9468.182600217431,
+      "required_pages_per_second_upper_estimate": 0.003964130336575067
     },
     {
       "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": 18864.9477707326,
-      "required_pages_per_second_upper_estimate": 0.0039290711185153175
+      "required_token_positions_per_second": 18936.365200434862,
+      "required_pages_per_second_upper_estimate": 0.003943945485718024
     },
     {
       "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": 28297.4216560989,
-      "required_pages_per_second_upper_estimate": 0.001696387220622065
+      "required_token_positions_per_second": 28404.547800652294,
+      "required_pages_per_second_upper_estimate": 0.0017028092694158893
     },
     {
       "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": 28297.4216560989,
-      "required_pages_per_second_upper_estimate": 0.002200000066557924
+      "required_token_positions_per_second": 28404.547800652294,
+      "required_pages_per_second_upper_estimate": 0.002208328653098838
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28297.4216560989,
-      "required_pages_per_second_upper_estimate": 0.3629689416002732
+      "required_token_positions_per_second": 28404.547800652294,
+      "required_pages_per_second_upper_estimate": 0.36434304075951174
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06644159145308492,
-    "producer_p95_seconds": 0.22380153400081326,
-    "gpu_update_seconds": 0.977050147395357,
-    "producer_batches_per_second": 15.050813475864889,
-    "gpu_batches_per_second": 1.0234889198531143,
-    "producer_headroom": 14.705399525013815,
+    "producer_batch_seconds": 0.11932395960936049,
+    "producer_p95_seconds": 0.22478175000105693,
+    "gpu_update_seconds": 0.9733652580579044,
+    "producer_batches_per_second": 8.380546566454655,
+    "gpu_batches_per_second": 1.0273635633916482,
+    "producer_headroom": 8.15733287132342,
     "sustainable": true,
-    "buffer_seconds": 60.577109138512135,
+    "buffer_seconds": 60.348645999590076,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 363854.91360346484,
-    "upload_bytes_per_second": 4841236.322648081,
+    "generation_bytes_per_second": 364453.85933892155,
+    "upload_bytes_per_second": 4769325.072746087,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.07479435041388,
+    "estimated_full_shards_per_hour": 78.20331416249977,
     "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": 1.2538779149999755,
-      "bytes": 7546399,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.2998433860002478,
       "bytes": 5962432,
@@ -562,6 +556,12 @@
       "bytes": 7370118,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.246054737999657,
+      "bytes": 1155172,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -145,3 +145,4 @@
 {"step": 734500, "loss": 0.552775502204895, "lr": 5e-05, "grad_norm": 0.23376881108886866, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107510795224, "tokens_per_second": 151102.58346296014, "data_wait_s": 1.6309999409713782e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 734600, "loss": 0.5546770691871643, "lr": 5e-05, "grad_norm": 0.23413498505396368, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107525465494, "tokens_per_second": 150669.92613693315, "data_wait_s": 2.1551000827457756e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 734700, "loss": 0.5249496698379517, "lr": 5e-05, "grad_norm": 0.21698075141996773, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107540170752, "tokens_per_second": 151096.98421900813, "data_wait_s": 1.6644999050186016e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 734800, "loss": 0.5689827799797058, "lr": 5e-05, "grad_norm": 0.22837274521446776, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107554848896, "tokens_per_second": 150016.09295800017, "data_wait_s": 2.1638999896822497e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:61fbcd422706dc4b5a89c6a402f15b2616f78b161083d5329491e0cbfc83ecd9
-size 313077
+oid sha256:9b8f80a5f64d2ac49fc71f3b2086d62748b296f415861944ed80b0b62141709e
+size 315195