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": 77237.11897227904,
-      "required_pages_per_second_upper_estimate": 0.005343443105309262
+      "required_token_positions_per_second": 76315.60363475882,
+      "required_pages_per_second_upper_estimate": 0.005279690536049442
     },
     {
       "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": 9654.63987153488,
-      "required_pages_per_second_upper_estimate": 0.0040421961024050945
+      "required_token_positions_per_second": 9539.450454344853,
+      "required_pages_per_second_upper_estimate": 0.0039939686988561904
     },
     {
       "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": 19309.27974306976,
-      "required_pages_per_second_upper_estimate": 0.004021613750531048
+      "required_token_positions_per_second": 19078.900908689706,
+      "required_pages_per_second_upper_estimate": 0.003973631914828105
     },
     {
       "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": 28963.91961460464,
-      "required_pages_per_second_upper_estimate": 0.0017363427555510272
+      "required_token_positions_per_second": 28618.35136303456,
+      "required_pages_per_second_upper_estimate": 0.0017156264665215688
     },
     {
       "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": 28963.91961460464,
-      "required_pages_per_second_upper_estimate": 0.002251817351217048
+      "required_token_positions_per_second": 28618.35136303456,
+      "required_pages_per_second_upper_estimate": 0.002224950939651575
     },
     {
       "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": 28963.91961460464,
-      "required_pages_per_second_upper_estimate": 0.3715180617822327
+      "required_token_positions_per_second": 28618.35136303456,
+      "required_pages_per_second_upper_estimate": 0.3670854832933718
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.10956271931263473,
-    "producer_p95_seconds": 0.30488543300452875,
-    "gpu_update_seconds": 0.9545669359632835,
-    "producer_batches_per_second": 9.127192226276556,
-    "gpu_batches_per_second": 1.0475954721717535,
-    "producer_headroom": 8.712515917384712,
+    "producer_batch_seconds": 0.4001273548278732,
+    "producer_p95_seconds": 0.3119410009967396,
+    "gpu_update_seconds": 0.9660933870464694,
+    "producer_batches_per_second": 2.499204285670946,
+    "gpu_batches_per_second": 1.0350966204801273,
+    "producer_headroom": 2.4144647332648965,
     "sustainable": true,
-    "buffer_seconds": 60.137716965686856,
+    "buffer_seconds": 60.86388338392757,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 338459.9641182156,
-    "upload_bytes_per_second": 4968193.415035955,
+    "generation_bytes_per_second": 313458.35171042365,
+    "upload_bytes_per_second": 4927547.988095205,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.6256293550477,
+    "estimated_full_shards_per_hour": 67.26086534008533,
     "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": 1.4557110550013022,
-      "bytes": 16684165,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 1.2760817240014148,
       "bytes": 7386323,
@@ -562,6 +556,12 @@
       "bytes": 931830,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.1493673139993916,
+      "bytes": 11660613,
+      "pages": 43,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -353,3 +353,4 @@
 {"step": 592400, "loss": 0.5743988156318665, "lr": 0.0003, "grad_norm": 0.1675508790609629, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86709705666, "tokens_per_second": 153783.93408761188, "data_wait_s": 1.056499604601413e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 592500, "loss": 0.6171957850456238, "lr": 0.0003, "grad_norm": 0.15644797548527117, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86724460025, "tokens_per_second": 153508.74861962476, "data_wait_s": 1.0397998266853392e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 592600, "loss": 0.6019682884216309, "lr": 0.0003, "grad_norm": 0.17789078100444394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86739171466, "tokens_per_second": 153163.23699273873, "data_wait_s": 1.3493998267222196e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 592700, "loss": 0.5813198685646057, "lr": 0.0003, "grad_norm": 0.14987911509467272, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86753634337, "tokens_per_second": 149733.81975582172, "data_wait_s": 8.442999387625605e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:98dce7195cdfa92a08aa7a329d725a1c70b8e5d4c62227d79d74ab7269151095
-size 757536
+oid sha256:3af401ce9199c2434c446b35fc3aad8ad78b221223f9cafff1964e36bd0f4720
+size 759654