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": 77008.6556382272,
-      "required_pages_per_second_upper_estimate": 0.005327637481751585
+      "required_token_positions_per_second": 76775.96427268238,
+      "required_pages_per_second_upper_estimate": 0.005311539353164841
     },
     {
       "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": 9626.0819547784,
-      "required_pages_per_second_upper_estimate": 0.004030239499016272
+      "required_token_positions_per_second": 9596.995534085298,
+      "required_pages_per_second_upper_estimate": 0.004018061622065603
     },
     {
       "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": 19252.1639095568,
-      "required_pages_per_second_upper_estimate": 0.004009718028656118
+      "required_token_positions_per_second": 19193.991068170595,
+      "required_pages_per_second_upper_estimate": 0.003997602159916364
     },
     {
       "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": 28878.245864335197,
-      "required_pages_per_second_upper_estimate": 0.0017312067450386162
+      "required_token_positions_per_second": 28790.986602255893,
+      "required_pages_per_second_upper_estimate": 0.0017259756855141405
     },
     {
       "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": 28878.245864335197,
-      "required_pages_per_second_upper_estimate": 0.002245156593972601
+      "required_token_positions_per_second": 28790.986602255893,
+      "required_pages_per_second_upper_estimate": 0.0022383725701588665
     },
     {
       "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": 28878.245864335197,
-      "required_pages_per_second_upper_estimate": 0.37041913090308226
+      "required_token_positions_per_second": 28790.986602255893,
+      "required_pages_per_second_upper_estimate": 0.36929986278082494
     }
   ],
   "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": 0.20756756571125834,
-    "producer_p95_seconds": 0.34355219500139356,
-    "gpu_update_seconds": 0.9573988714510338,
-    "producer_batches_per_second": 4.817708376418853,
-    "gpu_batches_per_second": 1.0444967398848088,
-    "producer_headroom": 4.612468562563602,
+    "producer_batch_seconds": 0.15116154790649716,
+    "producer_p95_seconds": 0.3303053409981658,
+    "gpu_update_seconds": 0.9603005406502348,
+    "producer_batches_per_second": 6.615439004491819,
+    "gpu_batches_per_second": 1.0413406612505747,
+    "producer_headroom": 6.352809652652145,
     "sustainable": true,
-    "buffer_seconds": 60.31612890141513,
+    "buffer_seconds": 60.49893406096479,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 366486.6185407579,
-    "upload_bytes_per_second": 5187338.264012252,
+    "generation_bytes_per_second": 366267.8961667483,
+    "upload_bytes_per_second": 5161192.984974496,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.63949696700146,
+    "estimated_full_shards_per_hour": 78.59256423713529,
     "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": 1.1941587969995453,
-      "bytes": 6022783,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 0.9699245899973903,
       "bytes": 5017188,
@@ -562,6 +556,12 @@
       "bytes": 5961993,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.7475975829947856,
+      "bytes": 6736460,
+      "pages": 41,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -332,3 +332,4 @@
 {"step": 637600, "loss": 0.5447022914886475, "lr": 0.0001, "grad_norm": 0.21281934312340497, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93328093606, "tokens_per_second": 152960.06271889122, "data_wait_s": 1.8478000129107386e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 637700, "loss": 0.5084192156791687, "lr": 0.0001, "grad_norm": 0.171511491103215, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93342791224, "tokens_per_second": 152764.1056242407, "data_wait_s": 1.5092999092303216e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 637800, "loss": 0.5254839062690735, "lr": 0.0001, "grad_norm": 0.1608861134915749, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93357526174, "tokens_per_second": 152757.9456056643, "data_wait_s": 1.2194002920296043e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 637900, "loss": 0.5674863457679749, "lr": 0.0001, "grad_norm": 0.16188142427010585, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93372038230, "tokens_per_second": 150246.5437670056, "data_wait_s": 1.0493000445421785e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7a0533551502a6a4b35791300cab120d372f4715949d20749fb90c6471c10866
-size 713058
+oid sha256:c5cec2512a1f915032e2f6424cd2e5d41d1dfdfe07435762f11d3e7307cf8545
+size 715176