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": 77218.40391757729,
-      "required_pages_per_second_upper_estimate": 0.005342148354400085
+      "required_token_positions_per_second": 76328.39762386477,
+      "required_pages_per_second_upper_estimate": 0.005280575653901936
     },
     {
       "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": 9652.300489697162,
-      "required_pages_per_second_upper_estimate": 0.004041216652081489
+      "required_token_positions_per_second": 9541.049702983097,
+      "required_pages_per_second_upper_estimate": 0.003994638270864915
     },
     {
       "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": 19304.600979394323,
-      "required_pages_per_second_upper_estimate": 0.00402063928744475
+      "required_token_positions_per_second": 19082.099405966193,
+      "required_pages_per_second_upper_estimate": 0.003974298077460752
     },
     {
       "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": 28956.901469091485,
-      "required_pages_per_second_upper_estimate": 0.0017359220284436012
+      "required_token_positions_per_second": 28623.149108949292,
+      "required_pages_per_second_upper_estimate": 0.0017159140840634233
     },
     {
       "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": 28956.901469091485,
-      "required_pages_per_second_upper_estimate": 0.00225127172127296
+      "required_token_positions_per_second": 28623.149108949292,
+      "required_pages_per_second_upper_estimate": 0.0022253239432933917
     },
     {
       "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": 28956.901469091485,
-      "required_pages_per_second_upper_estimate": 0.37142804054708745
+      "required_token_positions_per_second": 28623.149108949292,
+      "required_pages_per_second_upper_estimate": 0.3671470236265478
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09072189567968891,
-    "producer_p95_seconds": 0.14609609100034504,
-    "gpu_update_seconds": 0.9547982897794295,
-    "producer_batches_per_second": 11.022697359970213,
-    "gpu_batches_per_second": 1.0473416329966538,
-    "producer_headroom": 10.52445258805579,
+    "producer_batch_seconds": 0.39200497897654785,
+    "producer_p95_seconds": 0.1572655320001104,
+    "gpu_update_seconds": 0.9659314527120147,
+    "producer_batches_per_second": 2.5509880068636224,
+    "gpu_batches_per_second": 1.0352701500632702,
+    "producer_headroom": 2.464079551320706,
     "sustainable": true,
-    "buffer_seconds": 60.152292256104055,
+    "buffer_seconds": 60.85368152085693,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 503270.0271943424,
-    "upload_bytes_per_second": 5439527.993546694,
+    "generation_bytes_per_second": 448351.8188626412,
+    "upload_bytes_per_second": 5368089.655052602,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 107.99003230927185,
+    "estimated_full_shards_per_hour": 96.20586323174884,
     "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": [
@@ -358,6 +358,12 @@
       "bytes": 6078141,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.8920637669998541,
+      "bytes": 2452797,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -33,3 +33,4 @@
 {"step": 560400, "loss": 0.5924027562141418, "lr": 0.0003, "grad_norm": 0.15323526764925785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82024829166, "tokens_per_second": 153491.6710915769, "data_wait_s": 1.574399993842235e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 560500, "loss": 0.5484539866447449, "lr": 0.0003, "grad_norm": 0.15213750671061338, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82039572408, "tokens_per_second": 153307.04067925774, "data_wait_s": 1.7552999906911282e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 560600, "loss": 0.541042685508728, "lr": 0.0003, "grad_norm": 0.1687460609301299, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82054358655, "tokens_per_second": 153677.74438393157, "data_wait_s": 1.1931000244658208e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 560700, "loss": 0.6451371908187866, "lr": 0.0003, "grad_norm": 0.17166284654148692, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82069152100, "tokens_per_second": 153397.27259784646, "data_wait_s": 1.426699964213185e-05, "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:aa9e467dc2e7d99f27d80659c2382b2ca2df4b713b83716af384abd91aaafba1
-size 75078
+oid sha256:5069c351111db3d1ec96bcac68d312a32f4d1de841b204236d90cf1ce0bd8a18
+size 77196