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": 76228.29286063772,
-      "required_pages_per_second_upper_estimate": 0.005273650174106837
+      "required_token_positions_per_second": 76534.18210504572,
+      "required_pages_per_second_upper_estimate": 0.005294812301795815
     },
     {
       "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": 9528.536607579716,
-      "required_pages_per_second_upper_estimate": 0.003989399299122664
+      "required_token_positions_per_second": 9566.772763130715,
+      "required_pages_per_second_upper_estimate": 0.004005407978990144
     },
     {
       "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": 19057.07321515943,
-      "required_pages_per_second_upper_estimate": 0.003969085781900737
+      "required_token_positions_per_second": 19133.54552626143,
+      "required_pages_per_second_upper_estimate": 0.0039850129475928215
     },
     {
       "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": 28585.609822739145,
-      "required_pages_per_second_upper_estimate": 0.0017136636611742961
+      "required_token_positions_per_second": 28700.318289392148,
+      "required_pages_per_second_upper_estimate": 0.0017205402586004566
     },
     {
       "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": 28585.609822739145,
-      "required_pages_per_second_upper_estimate": 0.0022224054289084224
+      "required_token_positions_per_second": 28700.318289392148,
+      "required_pages_per_second_upper_estimate": 0.0022313235076414713
     },
     {
       "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": 28585.609822739145,
-      "required_pages_per_second_upper_estimate": 0.36666550996958924
+      "required_token_positions_per_second": 28700.318289392148,
+      "required_pages_per_second_upper_estimate": 0.36813686701545834
     }
   ],
   "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": 0.14458475428904194,
-    "producer_p95_seconds": 0.3285485329979565,
-    "gpu_update_seconds": 0.9671999363122454,
-    "producer_batches_per_second": 6.916358539441042,
-    "gpu_batches_per_second": 1.0339123923155074,
-    "producer_headroom": 6.68950153886003,
+    "producer_batch_seconds": 0.11885248057791387,
+    "producer_p95_seconds": 0.33022532500035595,
+    "gpu_update_seconds": 0.9633342641436456,
+    "producer_batches_per_second": 8.413791577067245,
+    "gpu_batches_per_second": 1.038061280721649,
+    "producer_headroom": 8.105293717552078,
     "sustainable": true,
-    "buffer_seconds": 60.93359598767146,
+    "buffer_seconds": 60.690058641049674,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 265507.1515534848,
-    "upload_bytes_per_second": 4773409.337941642,
+    "generation_bytes_per_second": 265518.2745536204,
+    "upload_bytes_per_second": 4734623.519998674,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.97165403321655,
+    "estimated_full_shards_per_hour": 56.97404077011546,
     "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": 1.1211934040002234,
-      "bytes": 5810769,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.2197048900015943,
       "bytes": 2772365,
@@ -562,6 +556,12 @@
       "bytes": 10703677,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.1966409989981912,
+      "bytes": 2669509,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -393,3 +393,4 @@
 {"step": 759300, "loss": 0.5393655896186829, "lr": 5e-05, "grad_norm": 0.23913106737895645, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111139486457, "tokens_per_second": 152788.36661414208, "data_wait_s": 7.922004442662e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 759400, "loss": 0.5784863829612732, "lr": 5e-05, "grad_norm": 0.2551689737982588, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111154092141, "tokens_per_second": 150167.8652714708, "data_wait_s": 1.6505997336935252e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 759500, "loss": 0.5208806395530701, "lr": 5e-05, "grad_norm": 0.2359393559232186, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111168881487, "tokens_per_second": 153367.73092413525, "data_wait_s": 1.2337004591245204e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 759600, "loss": 0.5625448226928711, "lr": 5e-05, "grad_norm": 0.24627748148452916, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111183611984, "tokens_per_second": 140438.81696518476, "data_wait_s": 2.0908002625219524e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:927cb0af6d8633ca3a491b93f71e595547b0476a96d8cf5d64a72748d59e74a1
-size 842256
+oid sha256:eaae43f321ca0175545af743ee4ecff305d1233fcdd11773f375fb8675d3c416
+size 844374