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": 76961.69158953922,
-      "required_pages_per_second_upper_estimate": 0.005324388399891745
+      "required_token_positions_per_second": 76006.51334289988,
+      "required_pages_per_second_upper_estimate": 0.005258306952470354
     },
     {
       "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": 9620.211448692402,
-      "required_pages_per_second_upper_estimate": 0.004027781640708172
+      "required_token_positions_per_second": 9500.814167862485,
+      "required_pages_per_second_upper_estimate": 0.003977792492523432
     },
     {
       "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": 19240.422897384804,
-      "required_pages_per_second_upper_estimate": 0.004007272685452004
+      "required_token_positions_per_second": 19001.62833572497,
+      "required_pages_per_second_upper_estimate": 0.003957538075694336
     },
     {
       "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": 28860.634346077208,
-      "required_pages_per_second_upper_estimate": 0.0017301509614102812
+      "required_token_positions_per_second": 28502.44250358746,
+      "required_pages_per_second_upper_estimate": 0.0017086779073802961
     },
     {
       "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": 28860.634346077208,
-      "required_pages_per_second_upper_estimate": 0.0022437873758929246
+      "required_token_positions_per_second": 28502.44250358746,
+      "required_pages_per_second_upper_estimate": 0.0022159395356587557
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28860.634346077208,
-      "required_pages_per_second_upper_estimate": 0.37019322925664383
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28502.44250358746,
+      "required_pages_per_second_upper_estimate": 0.36559872889762135
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08797849044543682,
-    "producer_p95_seconds": 0.23795810999945388,
-    "gpu_update_seconds": 0.9579831014268045,
-    "producer_batches_per_second": 11.366414619493701,
-    "gpu_batches_per_second": 1.043859749207075,
-    "producer_headroom": 10.88883312928555,
+    "producer_batch_seconds": 0.06746091640616214,
+    "producer_p95_seconds": 0.2505902529992454,
+    "gpu_update_seconds": 0.9700221304374208,
+    "producer_batches_per_second": 14.823397802355618,
+    "gpu_batches_per_second": 1.0309043150892454,
+    "producer_headroom": 14.37902391656238,
     "sustainable": true,
-    "buffer_seconds": 60.352935389888685,
+    "buffer_seconds": 60.14137208712009,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 269244.3843032545,
-    "upload_bytes_per_second": 3938075.587610757,
+    "generation_bytes_per_second": 250989.76970664598,
+    "upload_bytes_per_second": 4022377.564801318,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.773577182991275,
+    "estimated_full_shards_per_hour": 53.856561836238235,
     "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": 1.568186391999916,
-      "bytes": 1164036,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 2.5616059220010357,
       "bytes": 11807815,
@@ -562,6 +556,12 @@
       "bytes": 1189304,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 2.000728038001398,
+      "bytes": 11892571,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -195,3 +195,4 @@
 {"step": 819500, "loss": 0.5511067509651184, "lr": 3e-05, "grad_norm": 0.2806126738103894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119951999815, "tokens_per_second": 152030.24434130485, "data_wait_s": 9.433999366592616e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 819600, "loss": 0.5610944032669067, "lr": 3e-05, "grad_norm": 0.2856196797202541, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119966755033, "tokens_per_second": 153042.9228750421, "data_wait_s": 9.103001502808183e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 819700, "loss": 0.48045778274536133, "lr": 3e-05, "grad_norm": 0.2818042216552244, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119981363123, "tokens_per_second": 143385.86027960118, "data_wait_s": 1.123100082622841e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 819800, "loss": 0.5151752233505249, "lr": 3e-05, "grad_norm": 0.2876552129358285, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119995873760, "tokens_per_second": 150602.7573715653, "data_wait_s": 1.0370000381954014e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9b038fa41f2b4e8ddb7134989d909415f318ed1b9ed3c1413dc892228d86fb94
-size 419760
+oid sha256:40658c18ad27375b805abe2672ea2cfc11d276e3b6c2ec15a977b1baec25aaa1
+size 421878