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": 76828.99177773324,
-      "required_pages_per_second_upper_estimate": 0.005315207919005023
+      "required_token_positions_per_second": 76281.670678364,
+      "required_pages_per_second_upper_estimate": 0.005277342975390741
     },
     {
       "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": 9603.623972216656,
-      "required_pages_per_second_upper_estimate": 0.004020836810693674
+      "required_token_positions_per_second": 9535.2088347955,
+      "required_pages_per_second_upper_estimate": 0.003992192821325965
     },
     {
       "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": 19207.24794443331,
-      "required_pages_per_second_upper_estimate": 0.004000363217634551
+      "required_token_positions_per_second": 19070.417669591,
+      "required_pages_per_second_upper_estimate": 0.003971865079841904
     },
     {
       "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": 28810.871916649965,
-      "required_pages_per_second_upper_estimate": 0.001727167779748948
+      "required_token_positions_per_second": 28605.6265043865,
+      "required_pages_per_second_upper_estimate": 0.0017148636306753519
     },
     {
       "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": 28810.871916649965,
-      "required_pages_per_second_upper_estimate": 0.002239918565886751
+      "required_token_positions_per_second": 28605.6265043865,
+      "required_pages_per_second_upper_estimate": 0.0022239616378624265
     },
     {
       "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": 28810.871916649965,
-      "required_pages_per_second_upper_estimate": 0.369554930242685
+      "required_token_positions_per_second": 28605.6265043865,
+      "required_pages_per_second_upper_estimate": 0.3669222624695232
     }
   ],
   "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": 0.1317478384218589,
-    "producer_p95_seconds": 0.2823315400019055,
-    "gpu_update_seconds": 0.9596377395306133,
-    "producer_batches_per_second": 7.590257358135793,
-    "gpu_batches_per_second": 1.0420598928186475,
-    "producer_headroom": 7.283897413617037,
+    "producer_batch_seconds": 0.09999970239059053,
+    "producer_p95_seconds": 0.29144920999897295,
+    "gpu_update_seconds": 0.9665231417238964,
+    "producer_batches_per_second": 10.000029761029518,
+    "gpu_batches_per_second": 1.0346363753033312,
+    "producer_headroom": 9.665260181962715,
     "sustainable": true,
-    "buffer_seconds": 60.457177590428635,
+    "buffer_seconds": 60.890957928605474,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 307387.5672377791,
+    "generation_bytes_per_second": 318245.2878326479,
     "upload_bytes_per_second": 5554148.943592149,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.95821631288557,
+    "estimated_full_shards_per_hour": 68.28803039774492,
     "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": [

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

@@ -323,3 +323,4 @@
 {"step": 677300, "loss": 0.5256151556968689, "lr": 0.0001, "grad_norm": 0.19940030516919457, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99139353180, "tokens_per_second": 152680.23327248966, "data_wait_s": 1.342000177828595e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 677400, "loss": 0.5683329701423645, "lr": 0.0001, "grad_norm": 0.1972745044677394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99154059695, "tokens_per_second": 153635.39984273116, "data_wait_s": 1.0292002116329968e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 677500, "loss": 0.5617125630378723, "lr": 0.0001, "grad_norm": 0.1883111911669831, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99168767080, "tokens_per_second": 153016.44914447636, "data_wait_s": 1.4008997823111713e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 677600, "loss": 0.5918698906898499, "lr": 0.0001, "grad_norm": 0.2128557232345933, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99183354464, "tokens_per_second": 152053.96163570735, "data_wait_s": 8.975999662652612e-06, "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 sha256:312d0de577af23c407f5475a1af4cc0f3460bfbf34befcebec445a311a7bea9e
-size 693213
+oid sha256:46ab3ce6041cc256f3755acc667b04a14e12ba533cddfbaed3e825ca49166f69
+size 695331