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": 77004.21211613521,
-      "required_pages_per_second_upper_estimate": 0.00532733006858287
+      "required_token_positions_per_second": 76754.60610364293,
+      "required_pages_per_second_upper_estimate": 0.005310061745472911
     },
     {
       "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": 9625.526514516901,
-      "required_pages_per_second_upper_estimate": 0.004030006947777694
+      "required_token_positions_per_second": 9594.325762955366,
+      "required_pages_per_second_upper_estimate": 0.004016943844644662
     },
     {
       "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": 19251.053029033803,
-      "required_pages_per_second_upper_estimate": 0.004009486661539068
+      "required_token_positions_per_second": 19188.651525910733,
+      "required_pages_per_second_upper_estimate": 0.003996490074076834
     },
     {
       "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": 28876.579543550706,
-      "required_pages_per_second_upper_estimate": 0.0017311068516519105
+      "required_token_positions_per_second": 28782.977288866095,
+      "required_pages_per_second_upper_estimate": 0.0017254955394059356
     },
     {
       "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": 28876.579543550706,
-      "required_pages_per_second_upper_estimate": 0.0022450270448609775
+      "required_token_positions_per_second": 28782.977288866095,
+      "required_pages_per_second_upper_estimate": 0.002237749881272056
     },
     {
       "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": 28876.579543550706,
-      "required_pages_per_second_upper_estimate": 0.3703977571292147
+      "required_token_positions_per_second": 28782.977288866095,
+      "required_pages_per_second_upper_estimate": 0.3691971279083913
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.16506981752348793,
-    "producer_p95_seconds": 0.21246581000013975,
-    "gpu_update_seconds": 0.957454118078708,
-    "producer_batches_per_second": 6.0580426815926485,
-    "gpu_batches_per_second": 1.0444364707592124,
-    "producer_headroom": 5.80029791298746,
+    "producer_batch_seconds": 0.11011270389063554,
+    "producer_p95_seconds": 0.217977508000331,
+    "gpu_update_seconds": 0.9605677592878784,
+    "producer_batches_per_second": 9.081604253340329,
+    "gpu_batches_per_second": 1.0410509725429,
+    "producer_headroom": 8.723496248370386,
     "sustainable": true,
-    "buffer_seconds": 60.3196094389586,
+    "buffer_seconds": 60.51576883513634,
     "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": 374886.26101969654,
-    "upload_bytes_per_second": 5538224.539089838,
+    "generation_bytes_per_second": 421978.1280506367,
+    "upload_bytes_per_second": 5536575.035833069,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.44186470931217,
+    "estimated_full_shards_per_hour": 90.54668313159299,
     "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": [
-    {
-      "seconds": 2.084590124000897,
-      "bytes": 12097526,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 0.8632086769994203,
       "bytes": 2602865,
@@ -562,6 +556,12 @@
       "bytes": 5558514,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.273476110000047,
+      "bytes": 7479055,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -125,3 +125,4 @@
 {"step": 772500, "loss": 0.5680965185165405, "lr": 3e-05, "grad_norm": 0.25440132234263146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113071647572, "tokens_per_second": 153357.4423961609, "data_wait_s": 1.5044999599922448e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 772600, "loss": 0.5385637879371643, "lr": 3e-05, "grad_norm": 0.26398638208890207, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113086143598, "tokens_per_second": 151235.95749930415, "data_wait_s": 1.3949998901807703e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 772700, "loss": 0.5686334371566772, "lr": 3e-05, "grad_norm": 0.2466945608683629, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113100834187, "tokens_per_second": 152501.1983588463, "data_wait_s": 1.549300031911116e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 772800, "loss": 0.4973268210887909, "lr": 3e-05, "grad_norm": 0.24788275396210357, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113115549331, "tokens_per_second": 153642.2639014537, "data_wait_s": 1.1411000741645694e-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:f63188f4ae3bc4303e73484d1824d652c7f21a1c5e244eaa9044bf9d89425fa7
-size 270717
+oid sha256:0c2b0d68c1c7dea575d1a8b1cf5c3c1ee7f80fcd2f1e26c2137f93d4ca5fbf5b
+size 272835