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": 77194.85365535153,
-      "required_pages_per_second_upper_estimate": 0.00534051909261517
+      "required_token_positions_per_second": 76726.77479246422,
+      "required_pages_per_second_upper_estimate": 0.00530813631078803
     },
     {
       "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": 9649.356706918941,
-      "required_pages_per_second_upper_estimate": 0.0040399841517054225
+      "required_token_positions_per_second": 9590.846849058027,
+      "required_pages_per_second_upper_estimate": 0.004015487296043828
     },
     {
       "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": 19298.713413837882,
-      "required_pages_per_second_upper_estimate": 0.004019413062804888
+      "required_token_positions_per_second": 19181.693698116054,
+      "required_pages_per_second_upper_estimate": 0.00399504094203746
     },
     {
       "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": 28948.070120756824,
-      "required_pages_per_second_upper_estimate": 0.0017353926025956265
+      "required_token_positions_per_second": 28772.54054717408,
+      "required_pages_per_second_upper_estimate": 0.0017248698726774812
     },
     {
       "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": 28948.070120756824,
-      "required_pages_per_second_upper_estimate": 0.0022505851227849366
+      "required_token_positions_per_second": 28772.54054717408,
+      "required_pages_per_second_upper_estimate": 0.0022369384705118772
     },
     {
       "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": 28948.070120756824,
-      "required_pages_per_second_upper_estimate": 0.37131476149301346
+      "required_token_positions_per_second": 28772.54054717408,
+      "required_pages_per_second_upper_estimate": 0.3690632565920663
     }
   ],
   "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": 0.07126039638291104,
-    "producer_p95_seconds": 0.260216959999525,
-    "gpu_update_seconds": 0.9550895753902218,
-    "producer_batches_per_second": 14.033040100234555,
-    "gpu_batches_per_second": 1.047022212122281,
-    "producer_headroom": 13.402810310766975,
+    "producer_batch_seconds": 0.18783272229671866,
+    "producer_p95_seconds": 0.2882956439971167,
+    "gpu_update_seconds": 0.9609161886372064,
+    "producer_batches_per_second": 5.323885996926051,
+    "gpu_batches_per_second": 1.0406734862259144,
+    "producer_headroom": 5.115808240905174,
     "sustainable": true,
-    "buffer_seconds": 60.170643249583975,
+    "buffer_seconds": 60.537719884144,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 163,
+    "upload_pages_per_commit": 165,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 673591.9721320044,
-    "upload_bytes_per_second": 5355546.520191106,
+    "generation_bytes_per_second": 672379.5398290964,
+    "upload_bytes_per_second": 5380735.178534467,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.26857840046931,
+    "estimated_full_shards_per_hour": 72.13849852635703,
     "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": 1.0533692200006044,
-      "bytes": 7678729,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.9111081320006633,
       "bytes": 972942,
@@ -562,6 +556,12 @@
       "bytes": 1188235,
       "pages": 36,
       "shards": 1
+    },
+    {
+      "seconds": 1.200595816000714,
+      "bytes": 10503607,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -246,3 +246,4 @@
 {"step": 669600, "loss": 0.5544153451919556, "lr": 0.0001, "grad_norm": 0.2647767470857639, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98012503363, "tokens_per_second": 152129.6504761414, "data_wait_s": 2.0611001673387364e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 669700, "loss": 0.5682086944580078, "lr": 0.0001, "grad_norm": 0.1946624194911147, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98027024485, "tokens_per_second": 150728.8789047591, "data_wait_s": 9.079998562810943e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 669800, "loss": 0.595058262348175, "lr": 0.0001, "grad_norm": 0.19365635579819224, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98041666937, "tokens_per_second": 151575.4650209032, "data_wait_s": 1.4192002709023654e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 669900, "loss": 0.5402909517288208, "lr": 0.0001, "grad_norm": 0.1947490227177393, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98056439999, "tokens_per_second": 153219.01603945586, "data_wait_s": 1.0189996828557923e-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 sha256:27d4147d11d0e8d6d5e05013fd3bbfae1a305d2325efd70d1a2cb41d529695c4
-size 528561
+oid sha256:9a3d7662d91ef1c81c699e364b10c1d3b642e8bdc7716f042d381b3c0b158300
+size 530679