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": 76672.63208337966,
-      "required_pages_per_second_upper_estimate": 0.0053043905925451666
+      "required_token_positions_per_second": 76319.30442609121,
+      "required_pages_per_second_upper_estimate": 0.005279946565380836
     },
     {
       "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": 9584.079010422458,
-      "required_pages_per_second_upper_estimate": 0.0040126537433356575
+      "required_token_positions_per_second": 9539.913053261402,
+      "required_pages_per_second_upper_estimate": 0.003994162379362386
     },
     {
       "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": 19168.158020844916,
-      "required_pages_per_second_upper_estimate": 0.003992221817421697
+      "required_token_positions_per_second": 19079.826106522803,
+      "required_pages_per_second_upper_estimate": 0.003973824609137631
     },
     {
       "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": 28752.237031267374,
-      "required_pages_per_second_upper_estimate": 0.0017236527079005479
+      "required_token_positions_per_second": 28619.739159784207,
+      "required_pages_per_second_upper_estimate": 0.0017157096628176674
     },
     {
       "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": 28752.237031267374,
-      "required_pages_per_second_upper_estimate": 0.002235359961455859
+      "required_token_positions_per_second": 28619.739159784207,
+      "required_pages_per_second_upper_estimate": 0.0022250588347445875
     },
     {
       "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": 28752.237031267374,
-      "required_pages_per_second_upper_estimate": 0.368802824890232
+      "required_token_positions_per_second": 28619.739159784207,
+      "required_pages_per_second_upper_estimate": 0.36710328445997625
     }
   ],
   "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": 0.2520233501875566,
-    "producer_p95_seconds": 0.24090630999853602,
-    "gpu_update_seconds": 0.9615947437388422,
-    "producer_batches_per_second": 3.9678863059942526,
-    "gpu_batches_per_second": 1.0399391287350757,
-    "producer_headroom": 3.815498615597405,
+    "producer_batch_seconds": 0.07548425768750633,
+    "producer_p95_seconds": 0.2164306740014581,
+    "gpu_update_seconds": 0.9660465403140476,
+    "producer_batches_per_second": 13.247795376618159,
+    "gpu_batches_per_second": 1.0351468156750654,
+    "producer_headroom": 12.797986890370407,
     "sustainable": true,
-    "buffer_seconds": 60.58046885554706,
+    "buffer_seconds": 60.860932039785,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 369226.4804545935,
-    "upload_bytes_per_second": 5707976.077493876,
+    "generation_bytes_per_second": 371543.38340670237,
+    "upload_bytes_per_second": 5679427.305981269,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.22740755298952,
+    "estimated_full_shards_per_hour": 79.72456099177172,
     "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": 1.5156246349997673,
-      "bytes": 10630930,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.1701987970000118,
       "bytes": 7151255,
@@ -562,6 +556,12 @@
       "bytes": 6037243,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.2390309599995817,
+      "bytes": 6886865,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -139,3 +139,4 @@
 {"step": 773900, "loss": 0.5280709266662598, "lr": 3e-05, "grad_norm": 0.2468482830649305, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113276930616, "tokens_per_second": 151849.07020083806, "data_wait_s": 7.816001016180962e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 774000, "loss": 0.6277706623077393, "lr": 3e-05, "grad_norm": 0.3495191316206481, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113291394762, "tokens_per_second": 150799.23129527763, "data_wait_s": 1.5546000213362277e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 774100, "loss": 0.5496140718460083, "lr": 3e-05, "grad_norm": 0.24983376626042067, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113305966819, "tokens_per_second": 151470.92394922976, "data_wait_s": 1.2017000699415803e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 774200, "loss": 0.4660772383213043, "lr": 3e-05, "grad_norm": 0.28941746315473155, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113320583652, "tokens_per_second": 151429.46553302713, "data_wait_s": 1.1546000678208657e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:841aca34730d682532499a17984572ba0fdd5a51d724f7f082efa2b1ac0954f6
-size 300369
+oid sha256:87f954e4e385787550eba6cdbdd2c52c661cc5c3b6fd83b6fcca5b11033b7767
+size 302487