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": 77450.02450345711,
-      "required_pages_per_second_upper_estimate": 0.005358172403965054
+      "required_token_positions_per_second": 77482.50436410519,
+      "required_pages_per_second_upper_estimate": 0.005360419436088348
     },
     {
       "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": 9681.253062932139,
-      "required_pages_per_second_upper_estimate": 0.004053338489896494
+      "required_token_positions_per_second": 9685.313045513149,
+      "required_pages_per_second_upper_estimate": 0.004055038319821085
     },
     {
       "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": 19362.506125864278,
-      "required_pages_per_second_upper_estimate": 0.004032699402393039
+      "required_token_positions_per_second": 19370.626091026297,
+      "required_pages_per_second_upper_estimate": 0.004034390576998405
     },
     {
       "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": 29043.75918879642,
-      "required_pages_per_second_upper_estimate": 0.0017411290161158533
+      "required_token_positions_per_second": 29055.939136539448,
+      "required_pages_per_second_upper_estimate": 0.0017418591853853443
     },
     {
       "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": 29043.75918879642,
-      "required_pages_per_second_upper_estimate": 0.0022580245269332856
+      "required_token_positions_per_second": 29055.939136539448,
+      "required_pages_per_second_upper_estimate": 0.002258971464296378
     },
     {
       "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": 29043.75918879642,
-      "required_pages_per_second_upper_estimate": 0.372542158114909
+      "required_token_positions_per_second": 29055.939136539448,
+      "required_pages_per_second_upper_estimate": 0.3726983894067476
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0527305245078189,
-    "producer_p95_seconds": 0.1810333929997796,
-    "gpu_update_seconds": 0.9519428879807394,
-    "producer_batches_per_second": 18.96434767781837,
-    "gpu_batches_per_second": 1.0504831882521852,
-    "producer_headroom": 18.052975897093248,
+    "producer_batch_seconds": 0.049597507312501676,
+    "producer_p95_seconds": 0.13222645099995134,
+    "gpu_update_seconds": 0.9515438434144816,
+    "producer_batches_per_second": 20.162303595203813,
+    "gpu_batches_per_second": 1.0509237245565484,
+    "producer_headroom": 19.185315855069856,
     "sustainable": true,
-    "buffer_seconds": 60.924344830767325,
+    "buffer_seconds": 60.89880597852682,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -18,3 +18,4 @@
 {"step": 646800, "loss": 0.5722486972808838, "lr": 0.0001, "grad_norm": 0.17941181252062205, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94675145228, "tokens_per_second": 154065.66223758535, "data_wait_s": 1.634799991734326e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 646900, "loss": 0.5577962398529053, "lr": 0.0001, "grad_norm": 0.1942120188153206, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94689630493, "tokens_per_second": 151972.35427959543, "data_wait_s": 1.3982999917061534e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 647000, "loss": 0.578724205493927, "lr": 0.0001, "grad_norm": 0.17413717870327522, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94704061158, "tokens_per_second": 151499.552097465, "data_wait_s": 1.2176999916846398e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 647100, "loss": 0.541343629360199, "lr": 0.0001, "grad_norm": 0.18917820900379853, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94718643414, "tokens_per_second": 153102.9512978919, "data_wait_s": 7.906999599072151e-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:d0538a7927adb3db8ecf877ab898ca60fc008fe7f9f304372241ba21e7c3c2e9
-size 42525
+oid sha256:33b28cb10d59c19c0b24bbaf429d9f5acac18705f0f3a14a8844b8b71db6ceca
+size 44643