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": 76082.1497611201,
-      "required_pages_per_second_upper_estimate": 0.005263539655383499
+      "required_token_positions_per_second": 76593.86835184522,
+      "required_pages_per_second_upper_estimate": 0.005298941534840578
     },
     {
       "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": 9510.268720140013,
-      "required_pages_per_second_upper_estimate": 0.00398175091612853
+      "required_token_positions_per_second": 9574.233543980652,
+      "required_pages_per_second_upper_estimate": 0.004008531652133192
     },
     {
       "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": 19020.537440280026,
-      "required_pages_per_second_upper_estimate": 0.003961476343506533
+      "required_token_positions_per_second": 19148.467087961304,
+      "required_pages_per_second_upper_estimate": 0.00398812071538686
     },
     {
       "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": 28530.806160420037,
-      "required_pages_per_second_upper_estimate": 0.0017103782652986128
+      "required_token_positions_per_second": 28722.700631941956,
+      "required_pages_per_second_upper_estimate": 0.0017218820458604594
     },
     {
       "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": 28530.806160420037,
-      "required_pages_per_second_upper_estimate": 0.002218144685218947
+      "required_token_positions_per_second": 28722.700631941956,
+      "required_pages_per_second_upper_estimate": 0.0022330636363251982
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28530.806160420037,
-      "required_pages_per_second_upper_estimate": 0.3659625474329477
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28722.700631941956,
+      "required_pages_per_second_upper_estimate": 0.36842396367340025
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.1261753563517516,
-    "producer_p95_seconds": 0.25164124500224716,
-    "gpu_update_seconds": 0.9690577912360314,
-    "producer_batches_per_second": 7.925477913549145,
-    "gpu_batches_per_second": 1.0319301996679702,
-    "producer_headroom": 7.680246121393885,
+    "producer_batch_seconds": 0.09763320871090286,
+    "producer_p95_seconds": 0.23652416199911386,
+    "gpu_update_seconds": 0.9625835799455849,
+    "producer_batches_per_second": 10.242416624460775,
+    "gpu_batches_per_second": 1.038870827254845,
+    "producer_headroom": 9.859182061667626,
     "sustainable": true,
-    "buffer_seconds": 60.081583056633946,
+    "buffer_seconds": 60.642765536571844,
     "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": 419490.62088602234,
-    "upload_bytes_per_second": 4932871.526093504,
+    "generation_bytes_per_second": 428771.28761053365,
+    "upload_bytes_per_second": 4967562.173424197,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.01292200026991,
+    "estimated_full_shards_per_hour": 92.00433703648574,
     "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": 0.8000806340005511,
-      "bytes": 2307263,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.328747977000603,
       "bytes": 11901603,
@@ -562,6 +556,12 @@
       "bytes": 5156190,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.2706896670024435,
+      "bytes": 7639609,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -162,3 +162,4 @@
 {"step": 456400, "loss": 0.5894762873649597, "lr": 0.0003, "grad_norm": 0.1622674506318828, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66803094828, "tokens_per_second": 151932.17241486724, "data_wait_s": 1.1503001587698236e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 456500, "loss": 0.6160978674888611, "lr": 0.0003, "grad_norm": 0.1599696314445684, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66817832359, "tokens_per_second": 152691.77704402307, "data_wait_s": 1.2746000720653683e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 456600, "loss": 0.572414219379425, "lr": 0.0003, "grad_norm": 0.16415791665607574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66832543961, "tokens_per_second": 152526.55252933496, "data_wait_s": 1.2068998330505565e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 456700, "loss": 0.5616434216499329, "lr": 0.0003, "grad_norm": 0.18140111112097893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66847244892, "tokens_per_second": 151990.12268517324, "data_wait_s": 1.2554999557323754e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:503f2e875f5d46fa593dba52bc1b22b71bb64eb50fd54455b1049ce9a6ee9013
-size 349866
+oid sha256:b757f3bc6434596d3ea62d4c2f2e005b5ef6ad654558b4f82c245487e962c82a
+size 351984