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": 76868.15825035436,
-      "required_pages_per_second_upper_estimate": 0.005317917546459689
+      "required_token_positions_per_second": 77131.52601695101,
+      "required_pages_per_second_upper_estimate": 0.005336137939650256
     },
     {
       "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": 9608.519781294295,
-      "required_pages_per_second_upper_estimate": 0.0040228865837183625
+      "required_token_positions_per_second": 9641.440752118877,
+      "required_pages_per_second_upper_estimate": 0.00403666990673457
     },
     {
       "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": 19217.03956258859,
-      "required_pages_per_second_upper_estimate": 0.004002402553473961
+      "required_token_positions_per_second": 19282.881504237754,
+      "required_pages_per_second_upper_estimate": 0.004016115693550714
     },
     {
       "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": 28825.559343882884,
-      "required_pages_per_second_upper_estimate": 0.0017280482685851587
+      "required_token_positions_per_second": 28924.32225635663,
+      "required_pages_per_second_upper_estimate": 0.001733968954385724
     },
     {
       "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": 28825.559343882884,
-      "required_pages_per_second_upper_estimate": 0.0022410604487509457
+      "required_token_positions_per_second": 28924.32225635663,
+      "required_pages_per_second_upper_estimate": 0.0022487388307836378
     },
     {
       "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": 28825.559343882884,
-      "required_pages_per_second_upper_estimate": 0.36974332478909816
+      "required_token_positions_per_second": 28924.32225635663,
+      "required_pages_per_second_upper_estimate": 0.371010149386958
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05399430772659741,
-    "producer_p95_seconds": 0.13432903500051907,
-    "gpu_update_seconds": 0.9591487773113143,
-    "producer_batches_per_second": 18.520470807099606,
-    "gpu_batches_per_second": 1.0425911221022455,
-    "producer_headroom": 17.763886929859478,
+    "producer_batch_seconds": 0.06653700627341408,
+    "producer_p95_seconds": 0.15965528699962306,
+    "gpu_update_seconds": 0.9558737368141397,
+    "producer_batches_per_second": 15.029230438934942,
+    "gpu_batches_per_second": 1.0461632760545656,
+    "producer_headroom": 14.366046661105555,
     "sustainable": true,
-    "buffer_seconds": 60.4263729706128,
+    "buffer_seconds": 60.2200454192908,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -34,3 +34,4 @@
 {"step": 803400, "loss": 0.5424924492835999, "lr": 3e-05, "grad_norm": 0.28106106832345396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117594576659, "tokens_per_second": 154121.63457088725, "data_wait_s": 2.3081000108504668e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 803500, "loss": 0.5135341286659241, "lr": 3e-05, "grad_norm": 0.28913286282964196, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117609406982, "tokens_per_second": 154507.92294203243, "data_wait_s": 1.105800038203597e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 803600, "loss": 0.5231519937515259, "lr": 3e-05, "grad_norm": 0.2646122519530628, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117623837178, "tokens_per_second": 150324.62267700967, "data_wait_s": 1.383400012855418e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 803700, "loss": 0.5389294624328613, "lr": 3e-05, "grad_norm": 0.2732866946955584, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117638367567, "tokens_per_second": 150884.93780683746, "data_wait_s": 1.6381000023102388e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b9fb3de2eb92c9d3ab0c408f285af0e03cbec5352cfe239c4bd554a91479d818
-size 76413
+oid sha256:3f76d28777c2f2fd5364d395ed3e8ee96eff6c7621c2f3b1af4b88e6602a9307
+size 78531