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": 75688.88839590464,
-      "required_pages_per_second_upper_estimate": 0.0065605210144231854
+      "required_token_positions_per_second": 75351.31612105282,
+      "required_pages_per_second_upper_estimate": 0.006531261105208138
     },
     {
       "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": 9461.11104948808,
-      "required_pages_per_second_upper_estimate": 0.0037741786538567413
+      "required_token_positions_per_second": 9418.914515131602,
+      "required_pages_per_second_upper_estimate": 0.0037573458254075323
     },
     {
       "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": 18922.22209897616,
-      "required_pages_per_second_upper_estimate": 0.004377323062863228
+      "required_token_positions_per_second": 18837.829030263205,
+      "required_pages_per_second_upper_estimate": 0.004357800211683768
     },
     {
       "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": 28383.33314846424,
-      "required_pages_per_second_upper_estimate": 0.0017283960729012648
+      "required_token_positions_per_second": 28256.743545394806,
+      "required_pages_per_second_upper_estimate": 0.0017206874302386542
     },
     {
       "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": 28383.33314846424,
-      "required_pages_per_second_upper_estimate": 0.002185027555863642
+      "required_token_positions_per_second": 28256.743545394806,
+      "required_pages_per_second_upper_estimate": 0.0021752823377969074
     },
     {
       "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": 28383.33314846424,
-      "required_pages_per_second_upper_estimate": 0.35242600479859243
+      "required_token_positions_per_second": 28256.743545394806,
+      "required_pages_per_second_upper_estimate": 0.3508541855966178
     }
   ],
   "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": 0.18250267285934285,
-    "producer_p95_seconds": 0.19772663799994916,
-    "gpu_update_seconds": 0.9740927837961122,
-    "producer_batches_per_second": 5.479371804985633,
-    "gpu_batches_per_second": 1.0265962510295226,
-    "producer_headroom": 5.337416534972384,
+    "producer_batch_seconds": 0.49937937652339315,
+    "producer_p95_seconds": 0.24167164899881755,
+    "gpu_update_seconds": 0.9784566985074961,
+    "producer_batches_per_second": 2.0024855791239418,
+    "gpu_batches_per_second": 1.0220176340203562,
+    "producer_headroom": 1.9593454285584835,
     "sustainable": true,
-    "buffer_seconds": 60.393752595358954,
+    "buffer_seconds": 60.664315307464754,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 424428.5042793939,
-    "upload_bytes_per_second": 5170135.446716529,
+    "generation_bytes_per_second": 366770.8353441031,
+    "upload_bytes_per_second": 5231635.719438456,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.07247682844508,
+    "estimated_full_shards_per_hour": 78.70048327677078,
     "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": 1.2323853809998582,
-      "bytes": 7278265,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 0.9598245239999414,
       "bytes": 1062466,
@@ -562,6 +556,12 @@
       "bytes": 1010482,
       "pages": 32,
       "shards": 1
+    },
+    {
+      "seconds": 1.0687636729999213,
+      "bytes": 11353142,
+      "pages": 26,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -107,3 +107,4 @@
 {"step": 165700, "loss": 0.6078481357544661, "lr": 0.0003, "grad_norm": 0.13609650753847735, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24251122613, "tokens_per_second": 150290.87363514613, "data_wait_s": 1.2860000424552709e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 165800, "loss": 0.6403736751526594, "lr": 0.0003, "grad_norm": 0.14658243310423927, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24265691388, "tokens_per_second": 148917.27265474034, "data_wait_s": 9.196999599225819e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 165900, "loss": 0.5704298634082079, "lr": 0.0003, "grad_norm": 0.13297813866340955, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24280196012, "tokens_per_second": 147638.1694864609, "data_wait_s": 1.1963000360992737e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 166000, "loss": 0.5583077687770128, "lr": 0.0003, "grad_norm": 0.14372924377867893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24294726970, "tokens_per_second": 148583.72982506774, "data_wait_s": 1.3250000847619958e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bd01ee85491ea585e2f2b1c0d1b598c7c2ba12f179073184f67e502dfa3b83dc
-size 232593
+oid sha256:813295d908336e5348a70156e5039979620d6553a39e050116b15dfdda1adcbc
+size 234711