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": 76778.42813534019,
-      "required_pages_per_second_upper_estimate": 0.0053117098089004566
+      "required_token_positions_per_second": 76736.4631422582,
+      "required_pages_per_second_upper_estimate": 0.005308806573306846
     },
     {
       "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": 9597.303516917524,
-      "required_pages_per_second_upper_estimate": 0.004018190568046043
+      "required_token_positions_per_second": 9592.057892782275,
+      "required_pages_per_second_upper_estimate": 0.004015994334761703
     },
     {
       "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": 19194.607033835047,
-      "required_pages_per_second_upper_estimate": 0.003997730449320163
+      "required_token_positions_per_second": 19184.11578556455,
+      "required_pages_per_second_upper_estimate": 0.003995545398978241
     },
     {
       "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": 28791.910550752567,
-      "required_pages_per_second_upper_estimate": 0.0017260310748157285
+      "required_token_positions_per_second": 28776.173678346826,
+      "required_pages_per_second_upper_estimate": 0.001725087673343817
     },
     {
       "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": 28791.910550752567,
-      "required_pages_per_second_upper_estimate": 0.002238444403090463
+      "required_token_positions_per_second": 28776.173678346826,
+      "required_pages_per_second_upper_estimate": 0.002237220930480103
     },
     {
       "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": 28791.910550752567,
-      "required_pages_per_second_upper_estimate": 0.3693117142000817
+      "required_token_positions_per_second": 28776.173678346826,
+      "required_pages_per_second_upper_estimate": 0.3691098584977979
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,14 +158,14 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05388358271874871,
-    "producer_p95_seconds": 0.13456945500001893,
-    "gpu_update_seconds": 0.96026972406933,
-    "producer_batches_per_second": 18.558528396666013,
-    "gpu_batches_per_second": 1.0413740795266409,
-    "producer_headroom": 17.821192942599296,
+    "producer_batch_seconds": 0.04855519609374248,
+    "producer_p95_seconds": 0.13152493600000525,
+    "gpu_update_seconds": 0.9607948683185861,
+    "producer_batches_per_second": 20.59511814285257,
+    "gpu_batches_per_second": 1.040804892880021,
+    "producer_headroom": 19.78768382406776,
     "sustainable": true,
-    "buffer_seconds": 60.49699261636779,
+    "buffer_seconds": 60.53007670407092,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 71,

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

@@ -11,3 +11,4 @@
 {"step": 721100, "loss": 0.5602995157241821, "lr": 5e-05, "grad_norm": 0.21492721161542766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105548900762, "tokens_per_second": 152262.951680818, "data_wait_s": 1.646800001253723e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 721200, "loss": 0.5305733680725098, "lr": 5e-05, "grad_norm": 0.2155840424675001, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105563713753, "tokens_per_second": 153531.30067842387, "data_wait_s": 1.589700013937545e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 721300, "loss": 0.518498420715332, "lr": 5e-05, "grad_norm": 0.21344423893107728, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105578467144, "tokens_per_second": 153266.1153677966, "data_wait_s": 1.3526999964597053e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 721400, "loss": 0.5436562299728394, "lr": 5e-05, "grad_norm": 0.20138459009967297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105592944425, "tokens_per_second": 150507.00545782028, "data_wait_s": 1.8675999854167458e-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:f7a993027a0cff540bc3b8e9ca0501b73ccd2f7763d2bfb184a71d617b59ccd2
-size 27699
+oid sha256:99aa2845c341cb7b2808dd72f1f9389e1ff885163a6842a7fd66b5195ba18c1c
+size 29817