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": 76117.58822301876,
-      "required_pages_per_second_upper_estimate": 0.005265991370406208
+      "required_token_positions_per_second": 76329.79885507449,
+      "required_pages_per_second_upper_estimate": 0.005280672594328329
     },
     {
       "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": 9514.698527877345,
-      "required_pages_per_second_upper_estimate": 0.003983605584123246
+      "required_token_positions_per_second": 9541.224856884312,
+      "required_pages_per_second_upper_estimate": 0.003994711604145724
     },
     {
       "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": 19029.39705575469,
-      "required_pages_per_second_upper_estimate": 0.003963321567766134
+      "required_token_positions_per_second": 19082.449713768623,
+      "required_pages_per_second_upper_estimate": 0.00397437103733776
     },
     {
       "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": 28544.09558363203,
-      "required_pages_per_second_upper_estimate": 0.0017111749459283974
+      "required_token_positions_per_second": 28623.674570652933,
+      "required_pages_per_second_upper_estimate": 0.001715945584690223
     },
     {
       "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": 28544.09558363203,
-      "required_pages_per_second_upper_estimate": 0.0022191778794196843
+      "required_token_positions_per_second": 28623.674570652933,
+      "required_pages_per_second_upper_estimate": 0.002225364795629588
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28544.09558363203,
-      "required_pages_per_second_upper_estimate": 0.36613300988484027
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28623.674570652933,
+      "required_pages_per_second_upper_estimate": 0.3671537636850853
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10780322824217592,
-    "producer_p95_seconds": 0.1764766910000617,
-    "gpu_update_seconds": 0.9686066219542132,
-    "producer_batches_per_second": 9.276160058523827,
-    "gpu_batches_per_second": 1.0324108645700243,
-    "producer_headroom": 8.98495005899336,
+    "producer_batch_seconds": 0.08969080791403883,
+    "producer_p95_seconds": 0.18487950600137992,
+    "gpu_update_seconds": 0.9659137205377095,
+    "producer_batches_per_second": 11.149414563847131,
+    "gpu_batches_per_second": 1.0352891554778982,
+    "producer_headroom": 10.769372503182906,
     "sustainable": true,
-    "buffer_seconds": 60.05361056116122,
+    "buffer_seconds": 60.8525643938757,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 318289.43843778054,
-    "upload_bytes_per_second": 4885517.059048563,
+    "generation_bytes_per_second": 318043.58868907794,
+    "upload_bytes_per_second": 4906320.8998442935,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.2975040898329,
+    "estimated_full_shards_per_hour": 68.24475045685057,
     "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": [
@@ -370,6 +370,12 @@
       "bytes": 4930803,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.3428281450014765,
+      "bytes": 7443989,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -78,3 +78,4 @@
 {"step": 727800, "loss": 0.5481175184249878, "lr": 5e-05, "grad_norm": 0.23142053883470848, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106529601973, "tokens_per_second": 138000.22715871155, "data_wait_s": 2.3218999558594078e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 727900, "loss": 0.5419673919677734, "lr": 5e-05, "grad_norm": 0.22210347484844103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106544291727, "tokens_per_second": 152574.9030902761, "data_wait_s": 1.0648998795659281e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 728000, "loss": 0.5603241920471191, "lr": 5e-05, "grad_norm": 0.2217255008084377, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106558786331, "tokens_per_second": 150351.62201683602, "data_wait_s": 1.1181000445503742e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 728100, "loss": 0.5516948103904724, "lr": 5e-05, "grad_norm": 0.2244470638871388, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106573191688, "tokens_per_second": 149528.11418467318, "data_wait_s": 1.515400072094053e-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 sha256:074314190481af4f69efbe37b6287efbae8ef423ca51e299127efc2afaab0a11
-size 170388
+oid sha256:e81076e2acce8b4af2363dbdea1fdd9f72a01d7099066db700936def0dbd0dd0
+size 172506