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": 75547.17857198142,
-      "required_pages_per_second_upper_estimate": 0.006548237966045786
+      "required_token_positions_per_second": 75469.2353999124,
+      "required_pages_per_second_upper_estimate": 0.006541482049436007
     },
     {
       "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": 9443.397321497678,
-      "required_pages_per_second_upper_estimate": 0.0037671123829175353
+      "required_token_positions_per_second": 9433.65442498905,
+      "required_pages_per_second_upper_estimate": 0.00376322579583096
     },
     {
       "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": 18886.794642995355,
-      "required_pages_per_second_upper_estimate": 0.004369127544423999
+      "required_token_positions_per_second": 18867.3088499781,
+      "required_pages_per_second_upper_estimate": 0.004364619849147702
     },
     {
       "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": 28330.191964493035,
-      "required_pages_per_second_upper_estimate": 0.0017251600536076631
+      "required_token_positions_per_second": 28300.963274967155,
+      "required_pages_per_second_upper_estimate": 0.0017233801797666194
     },
     {
       "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": 28330.191964493035,
-      "required_pages_per_second_upper_estimate": 0.002180936600417326
+      "required_token_positions_per_second": 28300.963274967155,
+      "required_pages_per_second_upper_estimate": 0.002178686494987433
     },
     {
       "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": 28330.191964493035,
-      "required_pages_per_second_upper_estimate": 0.35176616914577197
+      "required_token_positions_per_second": 28300.963274967155,
+      "required_pages_per_second_upper_estimate": 0.35140324664399164
     }
   ],
   "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": 0.0773947382969169,
-    "producer_p95_seconds": 0.18576971699985734,
-    "gpu_update_seconds": 0.9759199667496767,
-    "producer_batches_per_second": 12.920775003639182,
-    "gpu_batches_per_second": 1.024674188530564,
-    "producer_headroom": 12.609642311931605,
+    "producer_batch_seconds": 0.2877085912656554,
+    "producer_p95_seconds": 0.1908782999998948,
+    "gpu_update_seconds": 0.9769278780858772,
+    "producer_batches_per_second": 3.4757390997637994,
+    "gpu_batches_per_second": 1.0236170166003744,
+    "producer_headroom": 3.3955464235123656,
     "sustainable": true,
-    "buffer_seconds": 60.50703793847996,
+    "buffer_seconds": 60.56952844132439,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 440008.454473314,
-    "upload_bytes_per_second": 6006781.799156426,
+    "generation_bytes_per_second": 411990.9548674681,
+    "upload_bytes_per_second": 5999888.675737051,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.41557145738187,
+    "estimated_full_shards_per_hour": 88.40366825597793,
     "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": [
@@ -490,6 +490,12 @@
       "bytes": 7389992,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.9817480370002158,
+      "bytes": 5514890,
+      "pages": 26,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -60,3 +60,4 @@
 {"step": 87900, "loss": 0.6350669208914042, "lr": 0.0003, "grad_norm": 0.13911254455738042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12863805001, "tokens_per_second": 150722.57719600401, "data_wait_s": 1.1763999282266013e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 88000, "loss": 0.5934349969029427, "lr": 0.0003, "grad_norm": 0.13046113793235078, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12878446667, "tokens_per_second": 148551.40109037986, "data_wait_s": 9.704999683890492e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 88100, "loss": 0.6505235023796558, "lr": 0.0003, "grad_norm": 0.14842643168683972, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12893146289, "tokens_per_second": 148335.49583834666, "data_wait_s": 1.1979000191786326e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 88200, "loss": 0.6165347024798393, "lr": 0.0003, "grad_norm": 0.13900520417190065, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12907718727, "tokens_per_second": 147939.0365733943, "data_wait_s": 9.487999705015682e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0a68983e6d46197c879e1dcdfc3b50099c532e29dde9bcd441c18958945b5764
-size 132264
+oid sha256:05141d1855b1c787fda8d85250a6d000579fc7f6bf092fd6770e1400e54ea3dc
+size 134382