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": 75420.99026320777,
-      "required_pages_per_second_upper_estimate": 0.006537300283262631
+      "required_token_positions_per_second": 74997.99440351961,
+      "required_pages_per_second_upper_estimate": 0.006500636074218066
     },
     {
       "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": 9427.623782900971,
-      "required_pages_per_second_upper_estimate": 0.003760820082535891
+      "required_token_positions_per_second": 9374.749300439951,
+      "required_pages_per_second_upper_estimate": 0.0037397276609382286
     },
     {
       "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": 18855.247565801943,
-      "required_pages_per_second_upper_estimate": 0.0043618296833248995
+      "required_token_positions_per_second": 18749.498600879902,
+      "required_pages_per_second_upper_estimate": 0.004337366521408403
     },
     {
       "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": 28282.871348702916,
-      "required_pages_per_second_upper_estimate": 0.0017222784763781284
+      "required_token_positions_per_second": 28124.247901319854,
+      "required_pages_per_second_upper_estimate": 0.00171261914066541
     },
     {
       "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": 28282.871348702916,
-      "required_pages_per_second_upper_estimate": 0.0021772937284219545
+      "required_token_positions_per_second": 28124.247901319854,
+      "required_pages_per_second_upper_estimate": 0.0021650824563446006
     },
     {
       "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": 28282.871348702916,
-      "required_pages_per_second_upper_estimate": 0.35117860546957197
+      "required_token_positions_per_second": 28124.247901319854,
+      "required_pages_per_second_upper_estimate": 0.3492090331316023
     }
   ],
   "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": 0.11168065442205943,
-    "producer_p95_seconds": 0.37822420799784595,
-    "gpu_update_seconds": 0.9775527972080518,
-    "producer_batches_per_second": 8.954102258577718,
-    "gpu_batches_per_second": 1.022962650054359,
-    "producer_headroom": 8.753107709359583,
+    "producer_batch_seconds": 0.2030690267501427,
+    "producer_p95_seconds": 0.38041583999438444,
+    "gpu_update_seconds": 0.9830662884571749,
+    "producer_batches_per_second": 4.924433903110225,
+    "gpu_batches_per_second": 1.0172254015234323,
+    "producer_headroom": 4.841044959883249,
     "sustainable": true,
-    "buffer_seconds": 60.60827342689921,
+    "buffer_seconds": 60.95010988434485,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 539938.7708872415,
-    "upload_bytes_per_second": 5256721.204694179,
+    "generation_bytes_per_second": 504696.7881723111,
+    "upload_bytes_per_second": 5304680.307190441,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 115.8582911011022,
+    "estimated_full_shards_per_hour": 108.29618200184822,
     "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": 1.058994241997425,
-      "bytes": 5472606,
-      "pages": 16,
-      "shards": 2
-    },
     {
       "seconds": 0.7334500290016877,
       "bytes": 861667,
@@ -562,6 +556,12 @@
       "bytes": 1158559,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.1566925789957168,
+      "bytes": 9706808,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -325,3 +325,4 @@
 {"step": 227500, "loss": 0.674255259335041, "lr": 0.0003, "grad_norm": 0.20765405607166754, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33299770332, "tokens_per_second": 149994.43463273006, "data_wait_s": 1.1252996046096087e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 227600, "loss": 0.5484527312219143, "lr": 0.0003, "grad_norm": 0.14529914166102892, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33314498363, "tokens_per_second": 149623.6974565787, "data_wait_s": 1.727099879644811e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 227700, "loss": 0.5897847693413496, "lr": 0.0003, "grad_norm": 0.1347731128257034, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33328951484, "tokens_per_second": 146795.19738615773, "data_wait_s": 1.5409001207444817e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 227800, "loss": 0.588342783972621, "lr": 0.0003, "grad_norm": 0.13741859228775813, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33343369061, "tokens_per_second": 146448.78013942097, "data_wait_s": 7.01400131219998e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:038fc98143512d949cfdaa63c68da952c7dd46f1a5b0d5829aec2aa7510c7dec
-size 697449
+oid sha256:97287449a5c86af4031ee64f533293af8f425fd75e169fc418c4ff9b57176dd4
+size 699567