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": 75604.10881771421,
-      "required_pages_per_second_upper_estimate": 0.0065531725354575155
+      "required_token_positions_per_second": 75388.22978788232,
+      "required_pages_per_second_upper_estimate": 0.006534460688292086
     },
     {
       "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": 9450.513602214276,
-      "required_pages_per_second_upper_estimate": 0.0037699511736932648
+      "required_token_positions_per_second": 9423.52872348529,
+      "required_pages_per_second_upper_estimate": 0.0037591865021083808
     },
     {
       "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": 18901.027204428552,
-      "required_pages_per_second_upper_estimate": 0.004372419997027044
+      "required_token_positions_per_second": 18847.05744697058,
+      "required_pages_per_second_upper_estimate": 0.004359935043474432
     },
     {
       "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": 28351.540806642828,
-      "required_pages_per_second_upper_estimate": 0.0017264600860858671
+      "required_token_positions_per_second": 28270.586170455867,
+      "required_pages_per_second_upper_estimate": 0.001721530373478232
     },
     {
       "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": 28351.540806642828,
-      "required_pages_per_second_upper_estimate": 0.0021825800933833947
+      "required_token_positions_per_second": 28270.586170455867,
+      "required_pages_per_second_upper_estimate": 0.002176347981392947
     },
     {
       "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": 28351.540806642828,
-      "required_pages_per_second_upper_estimate": 0.3520312503152939
+      "required_token_positions_per_second": 28270.586170455867,
+      "required_pages_per_second_upper_estimate": 0.35102606467159025
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07775888853907986,
-    "producer_p95_seconds": 0.31748362200232805,
-    "gpu_update_seconds": 0.9751850944736666,
-    "producer_batches_per_second": 12.860266122469364,
-    "gpu_batches_per_second": 1.025446354406931,
-    "producer_headroom": 12.54113983359678,
+    "producer_batch_seconds": 0.12648652779685676,
+    "producer_p95_seconds": 0.3140792300000612,
+    "gpu_update_seconds": 0.977977599519797,
+    "producer_batches_per_second": 7.905980323896996,
+    "gpu_batches_per_second": 1.0225183076698448,
+    "producer_headroom": 7.731871659015531,
     "sustainable": true,
-    "buffer_seconds": 60.46147585736733,
+    "buffer_seconds": 60.634611170227416,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 393802.57711689896,
-    "upload_bytes_per_second": 4093980.3109463276,
+    "generation_bytes_per_second": 392924.42285900575,
+    "upload_bytes_per_second": 4139138.89603614,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.50086579447009,
+    "estimated_full_shards_per_hour": 84.31243433311109,
     "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": 0.9874920929978543,
-      "bytes": 1116876,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.0614115569987916,
       "bytes": 7614945,
@@ -562,6 +556,12 @@
       "bytes": 9821385,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.015233548998367,
+      "bytes": 5854223,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -212,3 +212,4 @@
 {"step": 106200, "loss": 0.6398537121713161, "lr": 0.0003, "grad_norm": 0.14657011889388213, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15540802042, "tokens_per_second": 146805.74619366202, "data_wait_s": 1.0159998055314645e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 106300, "loss": 0.6330350320786238, "lr": 0.0003, "grad_norm": 0.1383405122334714, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15555285103, "tokens_per_second": 147163.03668029213, "data_wait_s": 1.1985001037828624e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 106400, "loss": 0.5892814230173826, "lr": 0.0003, "grad_norm": 0.1432565324174027, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15569722954, "tokens_per_second": 146102.891529372, "data_wait_s": 1.908599733724259e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 106500, "loss": 0.5803993083536625, "lr": 0.0003, "grad_norm": 0.12976601303436522, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15584311756, "tokens_per_second": 147872.0377700266, "data_wait_s": 1.5667999832658097e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:794f48eafd7f72295e5d678cd12c42218085f69d0b82a5824e27f95bc3a87f7e
-size 456549
+oid sha256:6f957388c51c6a34474cdceb0067ac4870fe8cf1600bc31438693a7a7db8ed94
+size 458667