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": 77327.83203541824,
-      "required_pages_per_second_upper_estimate": 0.005349718845500537
+      "required_token_positions_per_second": 77216.43524196895,
+      "required_pages_per_second_upper_estimate": 0.005342012156853543
     },
     {
       "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": 9665.97900442728,
-      "required_pages_per_second_upper_estimate": 0.004046943560559121
+      "required_token_positions_per_second": 9652.05440524612,
+      "required_pages_per_second_upper_estimate": 0.0040411136216606655
     },
     {
       "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": 19331.95800885456,
-      "required_pages_per_second_upper_estimate": 0.004026337035227935
+      "required_token_positions_per_second": 19304.10881049224,
+      "required_pages_per_second_upper_estimate": 0.004020536781641812
     },
     {
       "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": 28997.93701328184,
-      "required_pages_per_second_upper_estimate": 0.001738382046660165
+      "required_token_positions_per_second": 28956.163215738354,
+      "required_pages_per_second_upper_estimate": 0.0017358777712823286
     },
     {
       "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": 28997.93701328184,
-      "required_pages_per_second_upper_estimate": 0.002254462054337477
+      "required_token_positions_per_second": 28956.163215738354,
+      "required_pages_per_second_upper_estimate": 0.0022512143253219877
     },
     {
       "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": 28997.93701328184,
-      "required_pages_per_second_upper_estimate": 0.3719544004474268
+      "required_token_positions_per_second": 28956.163215738354,
+      "required_pages_per_second_upper_estimate": 0.37141857102574816
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.10506511681268194,
-    "producer_p95_seconds": 0.259846891000052,
-    "gpu_update_seconds": 0.9534471361647716,
-    "producer_batches_per_second": 9.517906897518383,
-    "gpu_batches_per_second": 1.0488258468345573,
-    "producer_headroom": 9.074821073721829,
+    "producer_batch_seconds": 0.1683315909220937,
+    "producer_p95_seconds": 0.27238346800004365,
+    "gpu_update_seconds": 0.9548226328884851,
+    "producer_batches_per_second": 5.940655550881203,
+    "gpu_batches_per_second": 1.0473149311247958,
+    "producer_headroom": 5.672272374175985,
     "sustainable": true,
-    "buffer_seconds": 60.067169578380614,
+    "buffer_seconds": 60.15382587197456,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 392149.50115223444,
+    "generation_bytes_per_second": 467853.81386138004,
     "upload_bytes_per_second": 5501182.353106481,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.14615417409206,
+    "estimated_full_shards_per_hour": 100.39053737526942,
     "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-v1/1790914434605211575-090606d4/train.jsonl

@@ -199,3 +199,4 @@
 {"step": 504900, "loss": 0.5882214903831482, "lr": 0.0003, "grad_norm": 0.15479383777678637, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73900679335, "tokens_per_second": 153322.55105851224, "data_wait_s": 8.228998922277242e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 505000, "loss": 0.5756577253341675, "lr": 0.0003, "grad_norm": 0.16870856170823553, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73915057124, "tokens_per_second": 149524.0963532161, "data_wait_s": 2.45060000452213e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 505100, "loss": 0.5841816067695618, "lr": 0.0003, "grad_norm": 0.17015466913908517, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73929691364, "tokens_per_second": 137519.0611919901, "data_wait_s": 1.6198999219341204e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 505200, "loss": 0.6202306747436523, "lr": 0.0003, "grad_norm": 0.16219923375316012, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73944182466, "tokens_per_second": 150991.73834246694, "data_wait_s": 1.4395998732652515e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c8d62ab1af4d3a275786043f1737bc6debc34cd0931790d5a1f5d281676a9687
-size 429015
+oid sha256:c491aa05acae62e3a351878a34af32f6f773c3ba3ae182b1cc5cd2c877c36b3a
+size 431133