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": 76539.5166911312,
-      "required_pages_per_second_upper_estimate": 0.005295181360839154
+      "required_token_positions_per_second": 76664.3098917704,
+      "required_pages_per_second_upper_estimate": 0.005303814844019493
     },
     {
       "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": 9567.4395863914,
-      "required_pages_per_second_upper_estimate": 0.00400568716396454
+      "required_token_positions_per_second": 9583.0387364713,
+      "required_pages_per_second_upper_estimate": 0.004012218202355699
     },
     {
       "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": 19134.8791727828,
-      "required_pages_per_second_upper_estimate": 0.003985290710992599
+      "required_token_positions_per_second": 19166.0774729426,
+      "required_pages_per_second_upper_estimate": 0.003991788494161382
     },
     {
       "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": 28702.3187591742,
-      "required_pages_per_second_upper_estimate": 0.00172066018370935
+      "required_token_positions_per_second": 28749.116209413904,
+      "required_pages_per_second_upper_estimate": 0.0017234656193956513
     },
     {
       "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": 28702.3187591742,
-      "required_pages_per_second_upper_estimate": 0.002231479035367889
+      "required_token_positions_per_second": 28749.116209413904,
+      "required_pages_per_second_upper_estimate": 0.0022351173312837964
     },
     {
       "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": 28702.3187591742,
-      "required_pages_per_second_upper_estimate": 0.3681625268938854
+      "required_token_positions_per_second": 28749.116209413904,
+      "required_pages_per_second_upper_estimate": 0.36876279433837306
     }
   ],
   "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": 0.21401135676562433,
-    "producer_p95_seconds": 0.20510156500131416,
-    "gpu_update_seconds": 0.9632671224920738,
-    "producer_batches_per_second": 4.6726492234482455,
-    "gpu_batches_per_second": 1.0381336356761501,
-    "producer_headroom": 4.501009371885815,
+    "producer_batch_seconds": 0.08929754014837954,
+    "producer_p95_seconds": 0.20719840200035833,
+    "gpu_update_seconds": 0.9616991283699586,
+    "producer_batches_per_second": 11.198516760241874,
+    "gpu_batches_per_second": 1.0398262517872505,
+    "producer_headroom": 10.769603807360983,
     "sustainable": true,
-    "buffer_seconds": 60.685828717000646,
+    "buffer_seconds": 60.587045087307395,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 395001.56328779605,
+    "generation_bytes_per_second": 460748.1520299648,
     "upload_bytes_per_second": 5204612.802286139,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.75814031577502,
+    "estimated_full_shards_per_hour": 98.86582775758941,
     "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-v1/1791039222312944374-63c10d34/train.jsonl

@@ -149,3 +149,4 @@
 {"step": 619300, "loss": 0.5463822484016418, "lr": 0.0001, "grad_norm": 0.1507567620039471, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90648449035, "tokens_per_second": 153014.174376825, "data_wait_s": 1.2853999578510411e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 619400, "loss": 0.5378100872039795, "lr": 0.0001, "grad_norm": 0.23478538433841434, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90663018080, "tokens_per_second": 151999.8661073202, "data_wait_s": 1.3214001228334382e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 619500, "loss": 0.5542622804641724, "lr": 0.0001, "grad_norm": 0.14574282078873887, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90677611079, "tokens_per_second": 151831.56581926986, "data_wait_s": 1.5812000128789805e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 619600, "loss": 0.5462628602981567, "lr": 0.0001, "grad_norm": 0.17916029485821122, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90692360414, "tokens_per_second": 154038.50334648308, "data_wait_s": 1.5442001313203946e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a52099ddc40719b59646fdea9b16fbeeb12e1dc61ecdfe6b9d64fb4f4de1b8e5
-size 322332
+oid sha256:d6f265f8400d2002e460bdfcfc1a78d724cea62ef6023a263dd4224822b4612b
+size 324450