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": 76983.58476642318,
-      "required_pages_per_second_upper_estimate": 0.005325903020667747
+      "required_token_positions_per_second": 76995.87710836295,
+      "required_pages_per_second_upper_estimate": 0.005326753433405303
     },
     {
       "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": 9622.948095802898,
-      "required_pages_per_second_upper_estimate": 0.004028927417705646
+      "required_token_positions_per_second": 9624.48463854537,
+      "required_pages_per_second_upper_estimate": 0.004029570736065258
     },
     {
       "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": 19245.896191605796,
-      "required_pages_per_second_upper_estimate": 0.004008412628297762
+      "required_token_positions_per_second": 19248.96927709074,
+      "required_pages_per_second_upper_estimate": 0.004009052670961561
     },
     {
       "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": 28868.844287408698,
-      "required_pages_per_second_upper_estimate": 0.0017306431348572493
+      "required_token_positions_per_second": 28873.45391563611,
+      "required_pages_per_second_upper_estimate": 0.0017309194750309879
     },
     {
       "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": 28868.844287408698,
-      "required_pages_per_second_upper_estimate": 0.0022444256627197324
+      "required_token_positions_per_second": 28873.45391563611,
+      "required_pages_per_second_upper_estimate": 0.002244784041038802
     },
     {
       "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": 28868.844287408698,
-      "required_pages_per_second_upper_estimate": 0.37029853756889597
+      "required_token_positions_per_second": 28873.45391563611,
+      "required_pages_per_second_upper_estimate": 0.37035766493036404
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.056895875437486865,
-    "producer_p95_seconds": 0.14698619299997517,
-    "gpu_update_seconds": 0.9577106629120871,
-    "producer_batches_per_second": 17.575966488093304,
-    "gpu_batches_per_second": 1.0441566944230576,
-    "producer_headroom": 16.832690516632464,
+    "producer_batch_seconds": 0.05167536717187815,
+    "producer_p95_seconds": 0.1416347690001203,
+    "gpu_update_seconds": 0.9575577650246936,
+    "producer_batches_per_second": 19.351579964083975,
+    "gpu_batches_per_second": 1.0443234199810514,
+    "producer_headroom": 18.530255660104892,
     "sustainable": true,
-    "buffer_seconds": 60.33577176346149,
+    "buffer_seconds": 60.326139196555694,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -34,3 +34,4 @@
 {"step": 723400, "loss": 0.5299516320228577, "lr": 5e-05, "grad_norm": 0.23368338442515782, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105885999701, "tokens_per_second": 150844.57008442422, "data_wait_s": 1.045400040311506e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 723500, "loss": 0.5432611107826233, "lr": 5e-05, "grad_norm": 0.22921308872240645, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105900622546, "tokens_per_second": 152058.41830552442, "data_wait_s": 1.6374000097130192e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 723600, "loss": 0.5428845882415771, "lr": 5e-05, "grad_norm": 0.21916657527971525, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105915173698, "tokens_per_second": 151386.29087988505, "data_wait_s": 1.0311000096407952e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 723700, "loss": 0.5550183653831482, "lr": 5e-05, "grad_norm": 0.22080920255256617, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105929663900, "tokens_per_second": 150692.5883327505, "data_wait_s": 9.963000138668576e-06, "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 sha256:92589a1b53aaa757108b0b953cd19799e9b7ca65576fc6f78a25f81e86d19615
-size 76413
+oid sha256:de31b11f20ef970a7bf55d4faee840993d56a09bd97f4674f5054c108a35d624
+size 78531