Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74984.74574866742,
-      "required_pages_per_second_upper_estimate": 0.006499487714393888
+      "required_token_positions_per_second": 75584.52931371324,
+      "required_pages_per_second_upper_estimate": 0.0065514754336745105
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9373.093218583428,
-      "required_pages_per_second_upper_estimate": 0.003739067025125031
+      "required_token_positions_per_second": 9448.066164214155,
+      "required_pages_per_second_upper_estimate": 0.0037689748540825573
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18746.186437166856,
-      "required_pages_per_second_upper_estimate": 0.0043366003106254805
+      "required_token_positions_per_second": 18896.13232842831,
+      "required_pages_per_second_upper_estimate": 0.004371287653611263
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28119.279655750284,
-      "required_pages_per_second_upper_estimate": 0.0017123166005765932
+      "required_token_positions_per_second": 28344.198492642467,
+      "required_pages_per_second_upper_estimate": 0.0017260129776853895
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28119.279655750284,
-      "required_pages_per_second_upper_estimate": 0.002164699987047671
+      "required_token_positions_per_second": 28344.198492642467,
+      "required_pages_per_second_upper_estimate": 0.002182014861726821
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28119.279655750284,
-      "required_pages_per_second_upper_estimate": 0.34914734414927656
+      "required_token_positions_per_second": 28344.198492642467,
+      "required_pages_per_second_upper_estimate": 0.35194008334855365
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07984135936721515,
-    "producer_p95_seconds": 0.2051098029996865,
-    "gpu_update_seconds": 0.9832399811972456,
-    "producer_batches_per_second": 12.524836850543716,
-    "gpu_batches_per_second": 1.017045705141431,
-    "producer_headroom": 12.314920349427172,
+    "producer_batch_seconds": 0.22576215596876636,
+    "producer_p95_seconds": 0.21035151599971869,
+    "gpu_update_seconds": 0.9754377075498118,
+    "producer_batches_per_second": 4.429440336042625,
+    "gpu_batches_per_second": 1.0251807903878207,
+    "producer_headroom": 4.320643127118086,
     "sustainable": true,
-    "buffer_seconds": 60.96087883422923,
+    "buffer_seconds": 60.477137868088334,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 379133.47978226194,
-    "upload_bytes_per_second": 5751403.84068372,
+    "generation_bytes_per_second": 376001.9736069638,
+    "upload_bytes_per_second": 5726084.5243813535,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.35321898556607,
+    "estimated_full_shards_per_hour": 80.68127065807997,
     "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": [
@@ -298,6 +298,12 @@
       "bytes": 1108574,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.2109743800001525,
+      "bytes": 6387518,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -71,3 +71,4 @@
 {"step": 122100, "loss": 0.6501970142126083, "lr": 0.0003, "grad_norm": 0.14229733065152866, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17867856303, "tokens_per_second": 149515.73548572362, "data_wait_s": 1.482800053054234e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 122200, "loss": 0.6173706008121371, "lr": 0.0003, "grad_norm": 0.13941618656200802, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17882556537, "tokens_per_second": 150057.172812769, "data_wait_s": 1.206800061481772e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 122300, "loss": 0.6414286438375711, "lr": 0.0003, "grad_norm": 0.14586759768145027, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17897353412, "tokens_per_second": 151554.98581152613, "data_wait_s": 1.3378000403463375e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 122400, "loss": 0.6381675526499748, "lr": 0.0003, "grad_norm": 0.17643904933300172, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17912108499, "tokens_per_second": 150145.00363960842, "data_wait_s": 1.4076999832468573e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:14284a45f24413112eb1d397e852d320b5dcba21004973734d1d6295679faaed
-size 155562
+oid sha256:632aabf2c6fd93250831eea2592f18c9ab2311a3d7fb20ec68971599d39d3dc1
+size 157680