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": 76401.26271160458,
-      "required_pages_per_second_upper_estimate": 0.005285616629742073
+      "required_token_positions_per_second": 76484.54845460207,
+      "required_pages_per_second_upper_estimate": 0.005291378530692188
     },
     {
       "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": 9550.157838950572,
-      "required_pages_per_second_upper_estimate": 0.003998451657194989
+      "required_token_positions_per_second": 9560.56855682526,
+      "required_pages_per_second_upper_estimate": 0.00400281040736861
     },
     {
       "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": 19100.315677901144,
-      "required_pages_per_second_upper_estimate": 0.003978092046509406
+      "required_token_positions_per_second": 19121.13711365052,
+      "required_pages_per_second_upper_estimate": 0.003982428602477815
     },
     {
       "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": 28650.473516851718,
-      "required_pages_per_second_upper_estimate": 0.0017175521405952186
+      "required_token_positions_per_second": 28681.705670475774,
+      "required_pages_per_second_upper_estimate": 0.0017194244605162436
     },
     {
       "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": 28650.473516851718,
-      "required_pages_per_second_upper_estimate": 0.0022274482958204367
+      "required_token_positions_per_second": 28681.705670475774,
+      "required_pages_per_second_upper_estimate": 0.0022298764583889817
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28650.473516851718,
-      "required_pages_per_second_upper_estimate": 0.3674975117924567
+      "required_token_positions_per_second": 28681.705670475774,
+      "required_pages_per_second_upper_estimate": 0.3678981243246723
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0819939674531156,
-    "producer_p95_seconds": 0.3008035729981202,
-    "gpu_update_seconds": 0.9650102286699701,
-    "producer_batches_per_second": 12.196019183628394,
-    "gpu_batches_per_second": 1.0362584460666855,
-    "producer_headroom": 11.769283261256579,
+    "producer_batch_seconds": 0.13949279121877112,
+    "producer_p95_seconds": 0.3155379529998754,
+    "gpu_update_seconds": 0.9639594073535227,
+    "producer_batches_per_second": 7.1688292367142274,
+    "gpu_batches_per_second": 1.0373880812527407,
+    "producer_headroom": 6.910460382441653,
     "sustainable": true,
-    "buffer_seconds": 60.79564440620812,
+    "buffer_seconds": 60.72944266327193,
     "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": 279355.3293464565,
-    "upload_bytes_per_second": 5612702.318401447,
+    "generation_bytes_per_second": 294192.00400419184,
+    "upload_bytes_per_second": 5627463.002855867,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.943150618508064,
+    "estimated_full_shards_per_hour": 63.12675561994855,
     "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": 1.1654375459984294,
-      "bytes": 2648981,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.4223995709980954,
       "bytes": 9800135,
@@ -562,6 +556,12 @@
       "bytes": 5863529,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.5651336049995734,
+      "bytes": 6044120,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -321,3 +321,4 @@
 {"step": 677100, "loss": 0.5368040800094604, "lr": 0.0001, "grad_norm": 0.20854331524786404, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99110019550, "tokens_per_second": 152624.3591450172, "data_wait_s": 9.710001904750243e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 677200, "loss": 0.5721763968467712, "lr": 0.0001, "grad_norm": 0.18616434732814036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99124691637, "tokens_per_second": 152216.9666889091, "data_wait_s": 9.499999578110874e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 677300, "loss": 0.5256151556968689, "lr": 0.0001, "grad_norm": 0.19940030516919457, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99139353180, "tokens_per_second": 152680.23327248966, "data_wait_s": 1.342000177828595e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 677400, "loss": 0.5683329701423645, "lr": 0.0001, "grad_norm": 0.1972745044677394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99154059695, "tokens_per_second": 153635.39984273116, "data_wait_s": 1.0292002116329968e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:98ce4ba78ec408dbc9ff1198f4dd72581006d2e9537178850524769bcfc264e7
-size 688977
+oid sha256:f6b7c09da65d7b012b701d0ccaf409fbf69ccdf5d98fc20c25014b7ad5ea6c56
+size 691095