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": 77382.40383159545,
-      "required_pages_per_second_upper_estimate": 0.005353494248984089
+      "required_token_positions_per_second": 77372.15359242466,
+      "required_pages_per_second_upper_estimate": 0.0053527851136544215
     },
     {
       "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": 9672.800478949432,
-      "required_pages_per_second_upper_estimate": 0.0040497995694929595
+      "required_token_positions_per_second": 9671.519199053082,
+      "required_pages_per_second_upper_estimate": 0.004049263124356524
     },
     {
       "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": 19345.600957898863,
-      "required_pages_per_second_upper_estimate": 0.004029178501725102
+      "required_token_positions_per_second": 19343.038398106164,
+      "required_pages_per_second_upper_estimate": 0.004028644788099529
     },
     {
       "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": 29018.401436848297,
-      "required_pages_per_second_upper_estimate": 0.0017396088576056086
+      "required_token_positions_per_second": 29014.557597159244,
+      "required_pages_per_second_upper_estimate": 0.0017393784252854533
     },
     {
       "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": 29018.401436848297,
-      "required_pages_per_second_upper_estimate": 0.0022560530732562826
+      "required_token_positions_per_second": 29014.557597159244,
+      "required_pages_per_second_upper_estimate": 0.0022557542316277227
     },
     {
       "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": 29018.401436848297,
-      "required_pages_per_second_upper_estimate": 0.37221689609995123
+      "required_token_positions_per_second": 29014.557597159244,
+      "required_pages_per_second_upper_estimate": 0.37216759145161354
     }
   ],
   "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": 0.06121479564068011,
-    "producer_p95_seconds": 0.19867395499932172,
-    "gpu_update_seconds": 0.9527747439902694,
-    "producer_batches_per_second": 16.33591992808113,
-    "gpu_batches_per_second": 1.0495660241915616,
-    "producer_headroom": 15.564451927323038,
+    "producer_batch_seconds": 0.26197179559379435,
+    "producer_p95_seconds": 0.21755887100152904,
+    "gpu_update_seconds": 0.95290096729605,
+    "producer_batches_per_second": 3.8172048167756585,
+    "gpu_batches_per_second": 1.0494269964250305,
+    "producer_headroom": 3.6374181622726662,
     "sustainable": true,
-    "buffer_seconds": 60.02480887138697,
+    "buffer_seconds": 60.03276093965115,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 547699.3069672075,
-    "upload_bytes_per_second": 5764454.871973159,
+    "generation_bytes_per_second": 502483.6277302379,
+    "upload_bytes_per_second": 5795112.190165907,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 117.5235214878289,
+    "estimated_full_shards_per_hour": 107.82128929071763,
     "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": 0.9610936170001878,
-      "bytes": 5798019,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.1097549460000664,
       "bytes": 5080154,
@@ -562,6 +556,12 @@
       "bytes": 6057143,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.2347321990000637,
+      "bytes": 9643472,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -129,3 +129,4 @@
 {"step": 537900, "loss": 0.5968229174613953, "lr": 0.0003, "grad_norm": 0.18029057161348447, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78731126659, "tokens_per_second": 153842.39206109423, "data_wait_s": 1.4650000593974255e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 538000, "loss": 0.5918370485305786, "lr": 0.0003, "grad_norm": 0.16774948382604338, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78745674851, "tokens_per_second": 152304.7358001179, "data_wait_s": 1.3949000276625156e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 538100, "loss": 0.5772097110748291, "lr": 0.0003, "grad_norm": 0.16119237883685197, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78760368063, "tokens_per_second": 153216.76422369707, "data_wait_s": 9.27600012801122e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 538200, "loss": 0.5897762775421143, "lr": 0.0003, "grad_norm": 0.15242153421038196, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78774895203, "tokens_per_second": 152190.3380791546, "data_wait_s": 9.842000508797355e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ff47b45996db28def193f85319220330f67e4feffa96e0c1aba7dab94a6f6359
-size 279189
+oid sha256:2df610e6460a166b7247ec9203dfbd99fbd90b5bdac5233d479e8cf3b4c92915
+size 281307