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": 76715.04637403593,
-      "required_pages_per_second_upper_estimate": 0.00530732491158748
+      "required_token_positions_per_second": 76041.62915785471,
+      "required_pages_per_second_upper_estimate": 0.0052607363460288456
     },
     {
       "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": 9589.38079675449,
-      "required_pages_per_second_upper_estimate": 0.004014873490558991
+      "required_token_positions_per_second": 9505.203644731839,
+      "required_pages_per_second_upper_estimate": 0.00397963027482593
     },
     {
       "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": 19178.76159350898,
-      "required_pages_per_second_upper_estimate": 0.003994430261972605
+      "required_token_positions_per_second": 19010.407289463677,
+      "required_pages_per_second_upper_estimate": 0.00395936650024153
     },
     {
       "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": 28768.142390263474,
-      "required_pages_per_second_upper_estimate": 0.001724606209886287
+      "required_token_positions_per_second": 28515.61093419552,
+      "required_pages_per_second_upper_estimate": 0.00170946733468822
     },
     {
       "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": 28768.142390263474,
-      "required_pages_per_second_upper_estimate": 0.002236596533157525
+      "required_token_positions_per_second": 28515.61093419552,
+      "required_pages_per_second_upper_estimate": 0.0022169633232167274
     },
     {
       "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": 28768.142390263474,
-      "required_pages_per_second_upper_estimate": 0.36900684175759
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28515.61093419552,
+      "required_pages_per_second_upper_estimate": 0.36576763938630236
     }
   ],
   "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": 0.26931410163280134,
-    "producer_p95_seconds": 0.3627840089975507,
-    "gpu_update_seconds": 0.9610630962865859,
-    "producer_batches_per_second": 3.713136422998966,
-    "gpu_batches_per_second": 1.040514409370062,
-    "producer_headroom": 3.5685583876218847,
+    "producer_batch_seconds": 0.1245757904842435,
+    "producer_p95_seconds": 0.36707382899476215,
+    "gpu_update_seconds": 0.9695741768886638,
+    "producer_batches_per_second": 8.02724185905512,
+    "gpu_batches_per_second": 1.0313806038120485,
+    "producer_headroom": 7.783006418179596,
     "sustainable": true,
-    "buffer_seconds": 60.54697506605491,
+    "buffer_seconds": 60.113598967097154,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 406060.42367540154,
-    "upload_bytes_per_second": 4582742.478588641,
+    "generation_bytes_per_second": 426565.46488532197,
+    "upload_bytes_per_second": 4644527.978541236,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.13111431786093,
+    "estimated_full_shards_per_hour": 91.53101882858033,
     "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": 1.527223689998209,
-      "bytes": 6243254,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 1.4163199819995498,
       "bytes": 7360891,
@@ -562,6 +556,12 @@
       "bytes": 9830750,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.8036188620026223,
+      "bytes": 13332219,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -394,3 +394,4 @@
 {"step": 643800, "loss": 0.5564897060394287, "lr": 0.0001, "grad_norm": 0.1720226383093094, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94235837802, "tokens_per_second": 152221.88280948895, "data_wait_s": 1.6689999029040337e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 643900, "loss": 0.5616509318351746, "lr": 0.0001, "grad_norm": 0.17307198984621422, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94250311988, "tokens_per_second": 150143.02719182655, "data_wait_s": 1.323899778071791e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 644000, "loss": 0.5245442986488342, "lr": 0.0001, "grad_norm": 0.17180469405553145, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94264816392, "tokens_per_second": 149160.67662374603, "data_wait_s": 9.87599923973903e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 644100, "loss": 0.5561860799789429, "lr": 0.0001, "grad_norm": 0.17385707982534387, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94279316177, "tokens_per_second": 150438.48340043137, "data_wait_s": 1.4274002751335502e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0bf39b501324c52f9a562e49ba43b6a3a2592dc49044a5ee75a1d446fdfae924
-size 845157
+oid sha256:f48533299630f1daf22266a1b06be9816db06d7f96199700b8727c98302ad81b
+size 847275