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": 77144.72567525468,
-      "required_pages_per_second_upper_estimate": 0.005337051122639131
+      "required_token_positions_per_second": 76451.66933918482,
+      "required_pages_per_second_upper_estimate": 0.0052891038771975644
     },
     {
       "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": 9643.090709406835,
-      "required_pages_per_second_upper_estimate": 0.004037360709396012
+      "required_token_positions_per_second": 9556.458667398103,
+      "required_pages_per_second_upper_estimate": 0.00400108968248971
     },
     {
       "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": 19286.18141881367,
-      "required_pages_per_second_upper_estimate": 0.004016802978732278
+      "required_token_positions_per_second": 19112.917334796206,
+      "required_pages_per_second_upper_estimate": 0.003980716639312607
     },
     {
       "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": 28929.272128220502,
-      "required_pages_per_second_upper_estimate": 0.0017342656916459467
+      "required_token_positions_per_second": 28669.37600219431,
+      "required_pages_per_second_upper_estimate": 0.001718685315729083
     },
     {
       "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": 28929.272128220502,
-      "required_pages_per_second_upper_estimate": 0.002249123661548868
+      "required_token_positions_per_second": 28669.37600219431,
+      "required_pages_per_second_upper_estimate": 0.0022289178809125772
     },
     {
       "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": 28929.272128220502,
-      "required_pages_per_second_upper_estimate": 0.3710736410284694
+      "required_token_positions_per_second": 28669.37600219431,
+      "required_pages_per_second_upper_estimate": 0.367739972578524
     }
   ],
   "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": 0.09359587150771631,
-    "producer_p95_seconds": 0.3283871699968586,
-    "gpu_update_seconds": 0.9557101843924161,
-    "producer_batches_per_second": 10.684231941977881,
-    "gpu_batches_per_second": 1.0463423078783458,
-    "producer_headroom": 10.211029279359021,
+    "producer_batch_seconds": 0.11146414806268012,
+    "producer_p95_seconds": 0.32697315900441026,
+    "gpu_update_seconds": 0.9643739716512792,
+    "producer_batches_per_second": 8.971494578128077,
+    "gpu_batches_per_second": 1.0369421297089956,
+    "producer_headroom": 8.651875857957291,
     "sustainable": true,
-    "buffer_seconds": 60.20974161672221,
+    "buffer_seconds": 60.755560214030595,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 322689.94032862724,
+    "generation_bytes_per_second": 317074.20821578515,
     "upload_bytes_per_second": 4624148.364742885,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.24174935716736,
+    "estimated_full_shards_per_hour": 68.03674397330443,
     "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-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -403,3 +403,4 @@
 {"step": 800300, "loss": 0.5441808700561523, "lr": 3e-05, "grad_norm": 0.27475873093016856, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117140665293, "tokens_per_second": 153062.38374156097, "data_wait_s": 1.7004000255838037e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 800400, "loss": 0.5579537153244019, "lr": 3e-05, "grad_norm": 0.2773674184649962, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117155369920, "tokens_per_second": 153523.41712702133, "data_wait_s": 1.5716002963017672e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 800500, "loss": 0.47471341490745544, "lr": 3e-05, "grad_norm": 0.26544661118539997, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117170013476, "tokens_per_second": 152721.78886785722, "data_wait_s": 1.53399960254319e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 800600, "loss": 0.5685527324676514, "lr": 3e-05, "grad_norm": 0.2735512879455177, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117184702810, "tokens_per_second": 152434.87473564062, "data_wait_s": 1.0414005373604596e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:117b6d6d770ed5bfea6742bd1118b2a861188012b6a83ad5cad23c5d8f5df9d0
-size 864219
+oid sha256:2ce7729a24cd4e7a218bb880528973e80697859bb4bac0f3245563054d081844
+size 866337